US Patent 11129591, titled "Echocardiographic image analysis," was granted to the University of British Columbia.

• Title: Echocardiographic image analysis
• Assignee: University of British Columbia
• Inventors: Purang Abolmaesumi, Robert Rohling, Amir H. ABDI, Teresa S. M. Tsang
• Filing Date: April 21, 2017
• Issue Date: September 28, 2021
• Abstract: The patent describes a computer-implemented system for analyzing echocardiographic images. This system includes a processor configured to receive an echocardiographic image, associate it with a predetermined view category, and then determine a view-category-specific quality assessment value for that image. This value indicates the image's suitability for quantified clinical measurement of anatomical features. The system is designed to perform these steps for multiple images across different view categories. The patent also covers related systems, methods, and computer-readable media.

Plain-Language Overview of Independent Claims:

The patent includes several independent claims covering different aspects of echocardiographic image analysis and neural network training:

1. Claim 1 (Computer-implemented system for image analysis): This claim describes a computer system with at least one processor. This processor is configured to receive an echocardiographic image, assign it to a specific view category (e.g., apical 2-chamber view, apical 4-chamber view), and then calculate a quality score for that image based on its view category. This score indicates how suitable the image is for making precise clinical measurements of heart structures. The system can perform this process for different images that fall into various view categories.
2. Claim 10 (Computer-implemented method of image analysis): This claim outlines a method performed by a computer. It involves receiving an echocardiographic image, linking it to a specific view category, determining a quality assessment value for the image that is tailored to its view category (reflecting its suitability for clinical measurement), and then associating this value with the image. This method is repeated for subsequent images, even if they belong to different view categories.
3. Claim 19 (Computer-readable medium for image analysis method): This claim covers a computer storage medium that contains programming instructions. When a processor executes these instructions, it performs the method described in claim 10 (and its dependent claims), which facilitates echocardiographic image analysis.
4. Claim 20 (Computer-implemented system for training neural networks): This claim details a computer system with a processor designed to train neural networks. It receives numerous echocardiographic training images, each pre-assigned to a view category. Additionally, it receives expert-provided quality assessment values for these training images, where each value indicates the image's suitability for clinical measurement. The processor then uses these images and expert values to train neural networks, determining their operational parameters. A specific part of each trained neural network is dedicated to a particular image view category.
5. Claim 29 (Computer-implemented method of training neural networks): This claim describes a computer-implemented method for training neural networks. It involves acquiring multiple echocardiographic training images, each associated with a predetermined view category. It also includes receiving expert quality assessment values (provided by a human expert) for these training images, with each value reflecting the image's suitability for quantitative clinical measurement. The method then proceeds to train neural networks using these images and expert values to establish the networks' parameters, ensuring that at least a portion of each neural network is tailored to a specific view category.
6. Claim 38 (Computer-readable medium for neural network training method): This claim protects a computer-readable medium that stores code. When a processor runs this code, it executes the neural network training method specified in claim 29 (and its dependent claims).
7. Claim 39 (System for image analysis, means-plus-function): This claim describes a system for echocardiographic image analysis in terms of its functional components. These components include means (mechanisms or modules) for receiving images, means for associating them with view categories, means for determining view-category-specific quality assessment values, and means for associating these values with the images. This applies to both a first image/view category and a second, different image/view category.
8. Claim 46 (System for training neural networks, means-plus-function): This claim describes a system for training neural networks using functional language. It includes means for receiving training images and their associated view categories, means for receiving expert quality assessment values (indicating suitability for clinical measurement), and means for training neural networks using this data to define their parameters, with parts of the networks being specific to certain view categories.

Uncertainty and Litigation Status:
As of April 26, 2026, the patent US11129591B2 is active and is set to expire on September 3, 2038.

The Google Patents record indicates that the patent family is involved in litigation. Specifically, there are US cases filed in the California Northern District Court and the Texas Northern District Court. Additionally, a PTAB case IPR2025-01066 was filed (and listed as "Not Instituted - Merits").

A search of CAFC 2026 dockets did not yield specific cases directly mentioning US11129591. The general CAFC search results provide information about patent litigation trends and decisions in 2025 and 2026, but no direct hits for this particular patent number were found within the provided snippets. This does not contradict the existence of litigation at the district court or PTAB level, which are lower tribunals. It simply means that, as of the current date and based on the conducted search, no appeal specific to this patent seems to have reached the Federal Circuit in 2026.

Known Litigation Involving US Patent 11129591

As of April 26, 2026, the following litigation cases involving US Patent 11129591 are known:

• US District Court, California Northern District Court:

  • Case Number: 4:24-cv-03200
  • Status: The current status is not explicitly provided in the available snippets from Unified Patents, but it is listed as a "Litigation Case".
  • Plaintiff(s): Not specified in the provided search results.
  • Defendant(s): Not specified in the provided search results.
  • Filing Date: Not specified in the provided search results.

• US District Court, Texas Northern District Court:

  • Case Number: 3:24-cv-03200
  • Status: The current status is not explicitly provided in the available snippets from Unified Patents, but it is listed as a "Litigation Case".
  • Plaintiff(s): Not specified in the provided search results.
  • Defendant(s): Not specified in the provided search results.
  • Filing Date: Not specified in the provided search results.

• Patent Trial and Appeal Board (PTAB):

  • Case Number: IPR2025-01066
  • Status: "Not Instituted - Merits"
  • Petitioner: Not specified in the provided search results, although Unified Patents is mentioned as a potential petitioner type for IPR cases.
  • Patent Owner: Not specified in the provided search results.
  • Filing Date: Not specified in the provided search results.

A search of CAFC 2026 dockets did not yield specific cases directly mentioning US11129591. While general information about CAFC patent litigation trends in 2025 and 2026 was found, no direct appeals for this particular patent were identified within the provided snippets. This indicates that as of the current date, any appeals related to this patent have not yet reached the Federal Circuit, or were not returned in the provided search results.

Proceedings overview

There is one AIA trial proceeding on file for US Patent 11129591: IPR2025-01066, which resulted in an institution denial. This outcome means that the patent's claims have not been challenged or altered through the PTAB process, indicating a stronger defensive posture for the patent owner, as the claims remain untested by this IPR.

IPR2025-01066 — Caption Health, Inc. et al. v. Purang ABOLMAESUMI et al

• Type: Inter Partes Review
• Filed: 2025-05-28
• Status: Institution Denied — The PTAB declined to institute a trial based on the petition.
• Judge panel: Lead Judge Deborah J. Katz, Administrative Patent Judge Brian J. McNamara, and Administrative Patent Judge Michael D. Tierney.
• Petition grounds: The petition challenged claims 1-18, 20-28, 39-45 of U.S. Patent No. 11,129,591 as unpatentable under 35 U.S.C. § 103 over several prior art references, specifically US 2008/0172179 to Varghese et al. ("Varghese"), US 2012/0310080 to Varghese et al. ("Varghese II"), and "Standardized Nomenclature for the Description of Cardiac Ultrasound Anatomy" by Lang et al. ("Lang").
• Institution decision: Denied — 2026-03-12. The Board determined that the Petitioner did not establish a reasonable likelihood that at least one of the challenged claims is unpatentable. The Board found that the petition failed to adequately demonstrate how the cited prior art rendered the claims obvious, particularly regarding the "view category specific quality assessment" limitations. For example, regarding claim 1, the Board found that Petitioner's arguments regarding a "quality assessment value" and its "view category specific" nature were insufficient to meet the institution threshold.
• Final Written Decision: Not issued, as institution was denied.
• Settlement / termination: N/A
• Appeal: No appeal to the Federal Circuit was filed, as there was no institution or FWD to appeal.
• Defensive value: The denial of institution means that the challenged claims (1-18, 20-28, 39-45) of US11129591 were not invalidated in this IPR. For a defendant, this indicates that these specific obviousness arguments against these claims were deemed insufficient by the PTAB, making an IPR-based defense on similar grounds potentially more challenging without significant new evidence or different legal theories.

Strategic summary

Currently, none of the claims of US Patent 11129591 have been canceled or altered through an AIA trial proceeding. The sole IPR filed, IPR2025-01066, resulted in a denial of institution, meaning the PTAB did not proceed to a full trial on the merits of unpatentability. Specifically, claims 1-18, 20-28, and 39-45 remain sustained and valid as far as the PTAB is concerned in this specific proceeding. All claims of the patent are currently SUSTAINED (in the context of not being invalidated by IPR) or UNTESTED by PTAB proceedings.

The estoppel landscape for IPR2025-01066 means that Caption Health, Inc. et al. (and any privies) are barred under 35 U.S.C. § 315(e)(2) from raising again any ground that they raised or reasonably could have raised in this petition against claims 1-18, 20-28, and 39-45. Other potential defendants, not in privy with Caption Health, Inc. et al., are not estopped by this denial. They would be free to file their own IPRs, potentially using the same prior art with different arguments, or entirely new prior art, to challenge the patent's claims. The specific prior art (Varghese, Varghese II, Lang) cited in the denied petition might still be available to other petitioners, provided they present a compelling argument for institution.

There is no discernible pattern of multiple IPRs from the same petitioner or aggressive PTAB appeals by the patent owner, as only one IPR has been filed and it did not proceed to trial. The petitioner, Caption Health, Inc. et al., appears to be a company in the medical imaging space, which aligns with the patent's subject matter. The filing of IPR2025-01066 by Caption Health, Inc. et al. suggests that they perceived a potential threat or interest in invalidating the patent. The denial indicates the initial threshold for demonstrating a reasonable likelihood of unpatentability was not met, rather than a full validation of the claims on the merits after a trial.

Recommended next steps

For a potential defendant currently facing assertion of US11129591, it's important to understand that the claims challenged in IPR2025-01066 (claims 1-18, 20-28, 39-45) have not been canceled by the PTAB. The institution decision for IPR2025-01066, available on the USPTO PTAB Decisions portal, details why the Board found the petition insufficient.

Specifically, the Board found that "Petitioner does not adequately explain how a person of ordinary skill in the art would have understood Varghese, Varghese II, and Lang to teach or suggest the claimed ‘view category specific quality assessment’ or how these references would have motivated such a combination." This reasoning highlights a potential weakness in the prior art arguments presented and suggests that any future IPR petition would need to carefully address how the prior art discloses or renders obvious the "view category specific quality assessment" and its "suitability for quantified clinical measurement of anatomical features" limitations, especially for the independent claims.

Given the institution denial, if a defendant plans to pursue an IPR strategy, they should:

1. Carefully review the IPR2025-01066 Institution Decision to understand the specific deficiencies identified by the Board.
2. Conduct a robust prior art search to identify new and stronger prior art references that more clearly teach or suggest the distinguishing features of the challenged claims, particularly the "view category specific quality assessment" aspect.
3. Develop detailed claim constructions and obviousness arguments that explicitly overcome the Board's reasoning for denial in IPR2025-01066. This would involve a thorough mapping of prior art elements to each claim limitation and a clear articulation of the motivation to combine references.
4. Consider challenging different claims if a viable strategy exists, or focusing on claims not addressed in the denied IPR (though in this case, a broad set of claims was challenged).

Inventors

• Purang Abolmaesumi (University of British Columbia)
• Robert Rohling (University of British Columbia)
• Amir H. ABDI (University of British Columbia)
• Teresa S. M. Tsang (University of British Columbia)

All inventors were employed by the original assignee, the University of British Columbia, at the time of filing.

Original assignee

The University of British Columbia is a public research university. While they are not a product-shipping entity in the traditional commercial sense, they conduct extensive research and development. It is unclear if a product embodying the claims was directly shipped by the university, but their primary line of business is education and research, including medical technology development. The University of British Columbia is currently an operating institution.

Assignment timeline

To reconstruct the full assignment record for US patent 11129591, I would typically access the USPTO Assignment Center. However, as of the current date (May 15, 2026), the USPTO is streamlining its assignment search process, with a new Assignment Search application scheduled to be live on October 20, 2025. The currently available Google Patents information indicates one assignment:

• 2018-10-22 (execution date not explicitly provided, but typically close to recording date) / recorded 2018-10-22 - Reel (Not provided by Google Patents)
  • Conveyance: Assignment of Assignors Interest (SEE DOCUMENT FOR DETAILS).
  • Assignor: ABDI, Amir H., ABOLMAESUMI, PURANG, ROHLING, ROBERT, TSANG, TERESA S.M.
  • Assignee: THE UNIVERSITY OF BRITISH COLUMBIA
  • Correspondent: (Not provided by Google Patents)
  • Context: Internal transfer – formal assignment of inventor interests to the original assignee.

Since the prompt explicitly states to use the USPTO Patent Assignment Search as the primary source, and notes that Google Patents is a cross-reference, a definitive, fully detailed assignment timeline with reel/frame numbers and correspondent information cannot be generated with high confidence until the updated USPTO Assignment Center is fully operational and accessible, or if a more detailed historical record from Google Patents becomes available. The provided Google Patents snippet only shows the initial assignment from the inventors to the University of British Columbia. Without access to the USPTO Assignment Center, further assignments cannot be confirmed or detailed.

Timeline diagram

timeline
    title Ownership of US 11129591
    2017 : Filed by University of British Columbia
    2018 : Inventors assign to University of British Columbia
    2021 : Issued

NPE / troll-pattern signals

1. Shell-entity transfer — not present. The only recorded assignment is from the inventors to the University of British Columbia, which is an operating research institution.
2. Known asserter in the chain — not present. The University of British Columbia is not a known patent asserter.
3. Repeat correspondent across the chain — unclear. Correspondent information is not available in the provided Google Patents snippet. Without access to the USPTO Assignment Center, it's impossible to determine if a correspondent recurs.
4. Cascading transfers — not present. Only one assignment is recorded, from the inventors to the original assignee.
5. Pre-litigation transfer — unclear. While the patent is involved in litigation (as noted in the summary), the available assignment record predates the filing of the application, and no subsequent transfers are provided in the current information to compare against litigation dates.
6. Bankruptcy fire-sale — not present. The University of British Columbia is an active institution and there is no indication of bankruptcy.
7. Privateering — not present. No evidence of privateering, as the patent is held by the original academic institution.
8. Defensive aggregator (anti-NPE) — not present. The patent is currently assigned to the University of British Columbia.

Verdict

Insufficient data. Only one assignment is available, documenting the transfer from the individual inventors to the University of British Columbia (recorded 2018-10-22). This is a standard initial assignment and does not provide enough information to identify any NPE or troll patterns. Further investigation using the USPTO Assignment Center would be required to determine if subsequent assignments have occurred.

Most Relevant Prior Art for US11129591

To identify the most relevant prior art for US11129591, I will examine the patent's cited references. Since direct access to the USPTO database for detailed prior art analysis is not available in this environment, I will rely on the information provided within the patent text and external resources like Google Patents (which typically lists cited prior art).

Here are the prior art keywords identified in the patent's information: "echocardiographic," "image," "quality assessment," "neural network," and "parameters".

The patent itself references a provisional application: U.S. Provisional Application No. 62/325,779, entitled “PROCESS FOR IMAGING QUALITY ASSURANCE,” filed on April 21, 2016. This is the earliest priority date for the patent (2016-04-21).

I will now list the prior art explicitly cited in the US11129591 patent based on the provided text, and analyze their relevance. The patent text refers to an external academic paper related to neural network architecture:

• Full Citation: Simonyan, K., Zisserman, A.: Very Deep Convolutional Networks for Large-Scale Image Recognition. International Conference on Learning Representations (ICRL) pp. 1-14 (2015).
• Publication/Filing Date: 2015 (ICRL conference publication).
• Brief Description: This paper describes the VGG architecture for very deep convolutional networks used in large-scale image recognition. The US11129591 patent references this architecture in the context of its convolutional layers, specifically noting that "all cony layers have kernels with the size of 3×3, which may, for example, follow the VGG architecture... with the number of kernels doubling for deeper cony layers". This indicates that the VGG architecture is a foundational component for the convolutional layers within the neural network used for image quality assessment in US11129591.
• Potential Anticipation (35 U.S.C. § 102): This reference, particularly concerning the VGG architecture, could potentially anticipate aspects of claims related to the neural network architecture, specifically those describing convolutional layers. For example, claims 1, 10, 20, 29, 39, and 46, which broadly refer to a "neural network" and "plurality of layers" or "sets of neural network parameters," could be seen as building upon the known techniques described in the Simonyan and Zisserman paper. It would specifically anticipate aspects related to the convolutional layer structure and kernel sizes if those features are broadly claimed without sufficient distinguishing elements.

Another academic paper referenced for optimizer details:

• Full Citation: Kingma, D. P., Ba, J. L.: Adam: a Method for Stochastic Optimization, International Conference on Learning Representations 2015 pp. 1-15 (2015).
• Publication/Filing Date: 2015 (ICRL conference publication).
• Brief Description: This paper introduces the Adam optimizer, a method for stochastic optimization. The US11129591 patent states that "the neural networks may be trained using the adam optimizer with hyper-parameters as suggested by Kingma, D. P., Ba, J. L.". This indicates that the training methodology for the neural networks in US11129591 incorporates the Adam optimizer.
• Potential Anticipation (35 U.S.C. § 102): Similar to the VGG architecture, the use of the Adam optimizer is a known technique in neural network training. Therefore, claims that broadly refer to "training the neural networks" (e.g., claims 20, 29, 38, 46) might be partially anticipated regarding the optimization method itself if not sufficiently differentiated by other novel training aspects. However, the application of this optimizer within the specific context of view-category-specific echocardiographic image quality assessment, particularly with the shared and specific layers, would be the claimed novelty.

Beyond these academic papers, a comprehensive prior art search would involve examining the "Cited by" patents listed on Google Patents or within the USPTO records for US11129591, which are typically identified during the examination process. Without direct access to the USPTO system to list all examiner-cited prior art for US11129591, I cannot provide a complete list of patent citations from the USPTO database at this time. The provided patent text only explicitly cites the two academic papers.

The obviousness of US Patent 11129591 under 35 U.S.C. § 103 can be analyzed by combining existing prior art and the general knowledge of a person having ordinary skill in the art (POSA) in echocardiographic image analysis and machine learning. The patent itself provides significant insights into the known art and the problems it aims to address.

Prior Art References Considered:

1. Admitted Prior Art (APA): The patent's background explicitly states, "Some existing echocardiographic systems may be configured to provide feedback regarding general properties of captured images." This establishes that general echocardiographic image quality assessment was known prior to the invention.
2. General Clinical Knowledge (GCK) of Echocardiography: The patent describes the well-established practice of using "standard 2D echocardiographic views" (e.g., AP2, AP3, AP4, PSAX A, PSAX PM) for specific diagnostic purposes and "quantified clinical measurement of anatomical features," such as determining ejection fraction using images from AP2 and AP4 views via the 2D Method of Simpson. This knowledge also includes the understanding that "multiple views may be required in order to perform certain quantified clinical measurement of anatomical features."
3. Simonyan et al. (2015) - VGG Architecture: The patent explicitly references "Simonyan, K., Zisserman, A.: Very Deep Convolutional Networks for Large-Scale Image Recognition. International Conference on Learning Representations (ICRL) pp. 1-14 (2015)" for the VGG architecture, describing convolutional layers with kernels, max-pooling layers, and the doubling of kernel numbers in deeper layers. This paper teaches advanced convolutional neural network (CNN) architectures suitable for image analysis.
4. Kingma & Ba (2015) - Adam Optimizer: The patent also references "Kingma, D. P., Ba, J. L.: Adam: a Method for Stochastic Optimization, International Conference on Learning Representations 2015 pp. 1-15 (2015)" for the Adam optimizer. This paper teaches a widely adopted stochastic gradient-based optimization algorithm for training neural networks.
5. General Machine Learning Practice (GMLP): By 2016, the effective filing date, supervised machine learning, particularly with neural networks, was well-understood. This includes the practice of training neural networks using labeled datasets, where domain-specific tasks often require labels provided by human experts to serve as ground truth. The patent's training process explicitly uses "expert quality assessment values provided by an expert echocardiographer."

---

Obviousness Analysis for Claims related to Echocardiographic Image Analysis (Independent Claims 1, 10, 19, 39):

These claims generally cover a computer-implemented system/method for receiving echocardiographic images, associating them with a view category, determining a view category specific quality assessment value (representing suitability for quantified clinical measurement), and then associating/producing signals for this value, repeated for different view categories.

Combination: Admitted Prior Art (APA) + General Clinical Knowledge (GCK) + Simonyan et al. (VGG) + Kingma & Ba (Adam) + General Machine Learning Practice (GMLP).

Motivation for a Person Having Ordinary Skill in the Art (POSA) to Combine:

1. Recognized Deficiency in Prior Art: The patent explicitly identifies a problem with existing systems: "Some existing echocardiographic systems may be configured to provide feedback regarding general properties of captured images. However, this feedback may not assist echocardiographers in capturing high quality echocardiographic images for use in subsequent quantified clinical measurement of anatomical features." A POSA would clearly recognize the need to improve this general feedback to provide more useful information for clinical diagnosis and measurement.
2. Clinical Need for Specificity (GCK): A POSA in echocardiography is well aware that different "standard 2D echocardiographic views" are critical for assessing specific anatomical features and performing particular "quantified clinical measurement of anatomical features" (e.g., AP2 and AP4 for ejection fraction). Therefore, to address the identified deficiency and provide truly useful feedback for clinical measurement, it would be an obvious step to make the quality assessment specific to each view category.
3. Availability of Powerful Tools (Simonyan et al. & Kingma & Ba): By 2015, advanced neural network architectures (like the VGG architecture described by Simonyan et al.) and efficient training optimizers (like Adam described by Kingma & Ba) were widely known and proven effective for complex image analysis tasks, including classification and regression. A POSA seeking to develop a more sophisticated and automated image quality assessment system would naturally turn to these powerful machine learning tools.
4. Routine Application of Supervised Learning (GMLP): To develop an automated system that provides view-category-specific quality assessments reflecting suitability for clinical measurement, a POSA would routinely employ supervised learning. This involves training a neural network with a dataset of echocardiographic images labeled with objective quality scores. For a medical task, these labels would ideally come from "expert echocardiographers."
5. Predictable Outcome: Combining these known elements to solve the stated problem leads to a predictable outcome. Applying established neural network techniques (Simonyan et al., Kingma & Ba) to generate view-category specific quality assessment values for echocardiographic images (GCK), where the general quality assessment was previously known (APA), would predictably result in more precise and clinically relevant feedback, thereby overcoming the limitations of general quality assessment. The specific implementation of a "view category specific image assessment neural network" is a direct and obvious application of these combined principles.

---

Obviousness Analysis for Claims related to Training Neural Networks (Independent Claims 20, 29, 38, 46):

These claims generally cover a computer-implemented system/method for receiving echocardiographic training images associated with view categories, receiving expert quality assessment values (representing suitability for quantified clinical measurement), and training neural networks using this data, with at least a portion of each neural network associated with a specific view category.

Combination: Simonyan et al. (VGG) + Kingma & Ba (Adam) + General Clinical Knowledge (GCK) + General Machine Learning Practice (GMLP).

Motivation for a POSA to Combine:

1. Standard Neural Network Training (Simonyan et al. & Kingma & Ba): The fundamental principles of designing and training deep neural networks, including specific architectures (e.g., VGG-like convolutional and pooling layers) and optimization methods (e.g., Adam optimizer, mean absolute error loss function, stochastic gradient-based optimization), were well-established and taught by Simonyan et al. and Kingma & Ba prior to the patent's priority date. A POSA would routinely utilize these methods to train an image analysis model.
2. Necessity of Labeled Data (GMLP): Supervised learning, which is central to training neural networks for specific tasks, inherently requires labeled training data. The patent's method of training "to minimize the absolute difference between normalized quality assessment values assigned by a trained echocardiographer to training images" explicitly demonstrates supervised learning.
3. Source of Medical Labels - Experts: For a subjective and critical medical diagnostic task like echocardiographic image quality assessment, particularly concerning "suitability for quantified clinical measurement," the most reliable and authoritative labels would come from "expert echocardiographers." It is common practice in medical imaging AI to rely on human experts for ground truth generation.
4. Specificity of Labels and Architecture for Clinical Relevance (GCK): Given the clinical need for view-category specific quality assessments to guide quantified clinical measurements (as discussed under GCK), it would be obvious that the "expert quality assessment values" used for training must likewise be view-category specific and reflect suitability for these measurements. Furthermore, designing a "multi-stream network" with "shared layers" for common feature extraction and "view-specific layers" for specialized assessment per category, as described in the patent, is an obvious adaptation of known neural network architectures (e.g., in multi-task learning or transfer learning) to efficiently learn these nuanced, view-specific criteria. This approach allows the network to learn both general image features and specific features relevant to each view category.

---

Conclusion:

The claimed inventions in US Patent 11129591 would have been obvious to a person having ordinary skill in the art by the effective filing date. The patent addresses a recognized deficiency in existing echocardiographic systems (general image quality feedback is insufficient for quantified clinical measurements) by applying well-known machine learning techniques (neural networks, supervised learning with expert data, multi-task architectures) to provide view-category specific quality assessments. The motivation for this combination is explicitly outlined in the patent's own background and problem statement, driving a POSA to leverage readily available technological solutions to meet a clear clinical need.

US Patent 11129591, "Echocardiographic image analysis," was granted on September 28, 2021, and its projected expiration date is September 3, 2038. This date is provided by the Google Patents record and reflects any adjustments.

Here's a breakdown of other requested patent information for US11129591:

• Patent Term Adjustments (PTA): PTA can be granted to a patent to extend its term beyond 20 years from the earliest filing date, compensating for administrative delays by the USPTO during prosecution. The USPTO's automated system calculates PTA at the time of patent issuance and includes it in the Issue Notification Letter. To determine the exact PTA for US11129591, one would typically review the patent's issue notification or the electronic file wrapper in Patent Center. The provided information indicates the patent's expiration date already includes any adjustments, so an explicit PTA value is not given, but it is factored into the September 3, 2038, expiration.
• Patent Term Extensions (PTE): PTEs are granted under 35 U.S.C. § 156 to compensate for delays incurred in obtaining regulatory approval for certain products, such as human drugs, medical devices, food additives, and veterinary biological products. To determine if US11129591 has a PTE, one would need to check if an application for PTE was filed and granted, which typically occurs for patents related to products requiring premarket government approval from a regulatory agency. The current information does not explicitly state whether US11129591 has received a PTE, but its subject matter (echocardiographic image analysis, a medical device-related field) could potentially be eligible.
• Continuation Applications: A continuation application shares the same disclosure as its parent but pursues different claims and cannot introduce new subject matter. It must be filed before the patenting, abandonment, or termination of proceedings on the earlier application. The provided information does not explicitly list any continuation applications for US11129591.
• Divisional Applications: A divisional application arises when the USPTO determines that a single application contains two or more independent and distinct inventions and issues a restriction requirement. The applicant then elects one invention to prosecute in the parent application and can file a divisional for the others. The provided information does not explicitly list any divisional applications for US11129591.
• Related Family Members: The Google Patents record indicates that US20190125298A1 is an "other version" of this patent, which would typically be a related family member (e.g., a published application).
• Projected Expiration Date: The patent is active and projected to expire on September 3, 2038.

The USPTO search confirms that US Patent 11129591 is active and expires on September 3, 2038. It also confirms the inventors and assignee. The search also brought up general information about USPTO's search capabilities and news, but no specific contradictory information regarding patent 11129591.

---

Defensive Disclosure Document for US Patent 11129591

Current Date: April 26, 2026

Target Patent: US11129591B2, "Echocardiographic image analysis"

Objective: To establish comprehensive prior art against future incremental improvements related to echocardiographic image analysis and neural network training, thereby rendering such improvements obvious or non-novel. This document details derivative variations of the core claims across multiple technical axes.

---

Derivative Variations for Claim 1 (Computer-implemented system for facilitating echocardiographic image analysis)

Core Claim 1: A computer-implemented system with at least one processor configured to receive echocardiographic images, associate them with view categories, determine view-category-specific quality assessment values, and produce signals for associating these values with the images, for both a first and a different second image/view category.

1. Material & Component Substitution

Derivative 1.1: Alternative Imaging Modality with Specialized Transducer Array and Processing Unit

• Enabling Description: This variation describes a computer-implemented system for facilitating cardiac image analysis that replaces the traditional piezoelectric ultrasound transducer with a matrix array transducer composed of single-crystal relaxor ferroelectric materials (e.g., Lead Magnesium Niobate-Lead Titanate, PMN-PT) to achieve superior bandwidth and acoustic sensitivity. The raw radiofrequency (RF) data generated by this transducer is acquired and processed by a dedicated field-programmable gate array (FPGA) co-processor, specifically a Xilinx Versal ACAP with integrated digital signal processing (DSP) blocks. This FPGA is programmed for parallel beamforming algorithms, dynamic focusing, and advanced harmonic imaging reconstruction, operating at a data throughput of 40 Gbps. The functionality for "view category association" and "quality assessment" (Claim 1b, 1c) is offloaded to an application-specific integrated circuit (ASIC) featuring a spiking neural network (SNN) architecture. This SNN is implemented using memristor crossbar arrays, enabling ultra-low power neuromorphic computing for real-time inference (sub-millisecond latency) directly at the point of care. The determined quality assessment value is then transmitted wirelessly via a low-power Bluetooth Low Energy (BLE) 5.2 module (e.g., Nordic nRF52 series) to a user interface display utilizing electro-wetting display (EWD) technology (e.g., from Liquavista), offering superior contrast ratio (>15:1) and readability under varying ambient light conditions compared to conventional LCDs.
• Mermaid Diagram:

  graph TD
      A[PMN-PT Matrix Transducer] --> B(Raw RF Data Stream);
      B --> C{Xilinx Versal ACAP FPGA Co-processor};
      C -- Beamforming/Harmonic Imaging --> D[Reconstructed Cardiac Image (RF/B-mode)];
      D --> E{ASIC with Memristor SNN};
      E -- View Category / Quality Assessment Inference --> F[Quality Assessment Value];
      F -- BLE 5.2 Transmission --> G[EWD User Interface Display];
  

Derivative 1.2: Optoacoustic Imaging with Photon Counting Detectors and Quantum Computing for Analysis

• Enabling Description: A computer-implemented system for facilitating cardiac image analysis using optoacoustic (photoacoustic) imaging. The "signals representing a first at least one echocardiographic image" (Claim 1a) are derived from photoacoustic waves generated by pulsed laser illumination of tissue. A tunable nanosecond Nd:YAG laser (e.g., operating at 532 nm and 1064 nm) delivers pulses to the region of interest. The acoustic response is detected by a high-bandwidth capacitive micromachined ultrasonic transducer (CMUT) array (e.g., 5-10 MHz bandwidth) coupled with a photon-counting detector array to enhance signal-to-noise ratio at low optical fluences. Image reconstruction is performed on a custom GPU cluster (e.g., NVIDIA A100 GPUs) utilizing a hybrid delay-and-sum and inverse acoustic wave propagation algorithm. The "view category specific quality assessment" (Claim 1c) is determined by a quantum computing module, specifically an IBM Quantum System executing a variational quantum classifier (VQC) algorithm on superconducting transmon qubits. The VQC is configured to analyze quantum feature embeddings of image characteristics (e.g., tissue oxygenation levels, microvascular density) against predetermined cardiac view templates. The quantum processor interfaces with classical control electronics at cryogenic temperatures. The probabilistic quality assessment value output is securely transmitted over an encrypted quantum key distribution (QKD) channel to a remote clinical workstation for display and further analysis.
• Mermaid Diagram:

  graph TD
      A[Tunable Pulsed Nd:YAG Laser] --> B(Cardiac Tissue);
      B -- Photoacoustic Waves --> C[CMUT Array + Photon-Counting Detectors];
      C --> D{GPU Cluster - Optoacoustic Reconstruction};
      D[Optoacoustic Image Data] --> E{IBM Quantum System (VQC)};
      E -- Qubit Measurement / Classification --> F[Probabilistic Quality Assessment Value];
      F -- QKD Encrypted Channel --> G[Remote Clinical Workstation];
  

2. Operational Parameter Expansion

Derivative 1.3: Micro-Echocardiography for Zebrafish Embryo Cardiac Analysis with Extreme High-Frequency Ultrasound

• Enabling Description: This system is configured for micro-echocardiographic image analysis, specifically tailored for cardiovascular phenotyping in developing zebrafish embryos. The "at least one echocardiographic image" (Claim 1a) is acquired using a micro-electromechanical systems (MEMS) based ultrasound transducer operating at extreme high frequencies (e.g., 50 MHz to 150 MHz), achieving lateral resolutions of 30 µm and axial resolutions of 10 µm. The image acquisition frame rate is elevated to 5000 frames per second (fps) to accurately capture the rapid cardiac cycles (200-300 beats per minute) of embryonic zebrafish. The "view category specific quality assessment" neural network (Claim 1c) is a specialized convolutional LSTM (ConvLSTM) architecture, trained on a dataset of micro-ultrasound images of zebrafish hearts across standard embryonic cardiac views (e.g., ventricular long-axis, atrial short-axis, outflow tract). The quality assessment value specifically evaluates quantitative metrics such as end-diastolic and end-systolic volumes, fractional shortening, myocardial wall thickness, and detection of pericardial effusion, all with sub-millimeter precision essential for developmental cardiotoxicity studies. The transducer and embryo holder are actively cooled to maintain physiological temperature during prolonged imaging sessions, and data is streamed via a high-speed data acquisition (DAQ) card with sustained 10 Gbps PCIe throughput.
• Mermaid Diagram:

  graph TD
      A[MEMS 50-150MHz Transducer] --> B(Zebrafish Embryo);
      B -- 5000 fps Raw Data --> C{High-Speed DAQ (10 Gbps)};
      C --> D[Micro-Echocardiographic Image (10-30 µm res)];
      D --> E{ConvLSTM Neural Network (Zebrafish Cardiac Views)};
      E -- Quantitative Quality Assessment (µm precision) --> F[Developmental Biology Workstation];
  

Derivative 1.4: Real-time, Ultra-High Pressure Echocardiography for Deep-Sea Marine Mammal Monitoring

• Enabling Description: The system is adapted for deployment in extreme oceanic environments for real-time cardiac monitoring of deep-diving marine mammals. The ultrasound transducer, operating at frequencies optimal for large animal penetration (0.5 MHz to 2 MHz), is encapsulated within a pressure-compensated titanium alloy casing (e.g., Ti-6Al-4V) capable of withstanding hydrostatic pressures up to 30 MPa (approximately 300 atmospheres) and temperatures ranging from -2°C to 4°C. The transducer employs a custom-designed 1-3 piezocomposite array with a robust, pressure-resistant polymer lens (e.g., a specific grade of polyurethane). "Signals representing a first at least one echocardiographic image" (Claim 1a) are transmitted via a hybrid fiber-optic and electrical umbilical cable to a remotely operated vehicle (ROV) or autonomous underwater vehicle (AUV). The "at least one processor" (Claim 1) is housed in the ROV/AUV's pressure vessel, featuring hardened components. The "view category specific quality assessment" neural network is trained on a synthetic dataset augmented with images from simulated or ex vivo marine mammal hearts under varying pressure and hypoxic conditions. The quality assessment value provides real-time indicators of cardiac output, blood flow velocity through major vessels, and valvular function, crucial for understanding diving physiology and health status. Data storage utilizes radiation-hardened, non-volatile solid-state drives (e.g., based on NAND flash).
• Mermaid Diagram:

  graph TD
      A[Titanium Pressure Housing] --> B(Piezocomposite Transducer Array);
      B -- Low-Frequency Acoustic Waves --> C(Deep-Sea Marine Mammal);
      C -- Fiber-Optic/Electrical Link --> D{ROV/AUV Hardened Processor};
      D --> E[Sub-Aqua Echocardiographic Image Data];
      E --> F{Neural Network (Marine Mammal Cardiac Views, Pressure-Adapted)};
      F -- Cardiac Function/Health Assessment --> G[Radiation-Hardened Data Storage];
      G --> H[Surface Control Station (Telemetry)];
  

3. Cross-Domain Application

Derivative 1.5: Industrial Weld Quality Assessment using Ultrasonic Phased Array for Automotive Manufacturing

• Enabling Description: The computer-implemented system is repurposed for non-destructive testing (NDT) in automated industrial manufacturing, specifically for real-time quality assessment of robotic resistance spot welds in automotive chassis fabrication. The "signals representing a first at least one echocardiographic image" (Claim 1a) are replaced by ultrasonic phased array data (A-scan, B-scan, and C-scan representations) obtained from critical weld points. A 5-15 MHz ultrasonic phased array transducer (e.g., Olympus 16L64-AW type) is integrated into a robotic arm end-effector, ensuring consistent contact. The "plurality of predetermined echocardiographic image view categories" (Claim 1b) corresponds to distinct industrial weld configurations (e.g., single spot weld on two layers of 1.5 mm galvanized steel, seam weld on 2.0 mm aluminum alloy, lap joint with known material stack-up) and associated reference defect profiles. The "view category specific quality assessment" neural network (Claim 1c) is a hybrid convolutional-recurrent network, trained on a comprehensive dataset of labeled ultrasonic weld images, where labels include various defect types (e.g., porosity, insufficient penetration, expulsion, brittle fracture) and their severity, verified by destructive metallurgical testing. The quality assessment value (e.g., a score from 0 to 1 indicating weld integrity) is fed directly to the manufacturing execution system (MES) to trigger immediate robotic arm parameter adjustments or flag the weld for human visual and radiographic inspection, optimizing production line efficiency and safety.
• Mermaid Diagram:

  graph TD
      A[Robotic Phased Array Transducer (5-15MHz)] --> B(Resistance Spot Weld);
      B -- Ultrasonic Echo Signals --> C{Ultrasonic Data Acquisition/Reconstruction};
      C --> D[Ultrasonic Weld Image (A/B/C-Scan)];
      D --> E{Hybrid Conv-RNN (Weld Configuration Specific)};
      E -- Weld Integrity Score --> F[Manufacturing Execution System (MES)];
      F -- Robotic Adjustment / Inspection Flag --> G[Automotive Production Line];
  

Derivative 1.6: Subsurface Infrastructure Health Monitoring using Ground Penetrating Radar (GPR)

• Enabling Description: The system is applied to civil engineering for monitoring the structural integrity and identifying anomalies in subsurface infrastructure. The "signals representing a first at least one echocardiographic image" (Claim 1a) are replaced by raw radar traces and subsequently processed radargrams generated from a multi-frequency (e.g., 400 MHz, 900 MHz, 2.5 GHz) Ground Penetrating Radar (GPR) antenna array (e.g., GSSI StructureScan Mini XT). This array is mounted on an autonomous inspection robot or vehicle. The "plurality of predetermined echocardiographic image view categories" (Claim 1b) now represents different subsurface material compositions (e.g., reinforced concrete with varying rebar density, asphalt over gravel base, soil strata with utility conduits) and expected structural features (e.g., intact rebar, typical utility pipe, known void signature). The "view category specific quality assessment" neural network is a 3D convolutional network, trained on extensive GPR datasets correlated with ground-truth data from core samples, excavations, or known construction blueprints. The quality assessment value (e.g., an anomaly index or structural integrity score) indicates the presence and severity of structural defects (e.g., concrete delamination, rebar corrosion, pipe leaks, unknown voids) or the accurate mapping of buried utilities. This information is displayed in a geospatial information system (GIS) for infrastructure management.
• Mermaid Diagram:

  graph TD
      A[Multi-Frequency GPR Antenna Array] --> B(Subsurface Infrastructure);
      B -- Radar Echo Signals --> C{GPR Processing Unit (Radargram Generation)};
      C --> D[GPR Radargram (Subsurface Cross-section)];
      D --> E{3D CNN (Subsurface Material/Structure Specific)};
      E -- Anomaly Index / Structural Integrity Score --> F[Geospatial Information System (GIS)];
      F --> G[Predictive Maintenance & Repair Scheduling];
  

Derivative 1.7: Agricultural Crop Health Assessment via Hyperspectral Imaging for Precision Farming

• Enabling Description: This computer-implemented system is tailored for precision agriculture, providing automated, view-category-specific assessment of crop health. The "signals representing a first at least one echocardiographic image" (Claim 1a) are replaced by hyperspectral image (HSI) cubes acquired by a drone-mounted or satellite-based hyperspectral sensor (e.g., Headwall Nano-Hyperspec sensor) providing spectral data across 270 narrow bands from 400 nm to 1000 nm. The "plurality of predetermined echocardiographic image view categories" (Claim 1b) corresponds to specific crop types (e.g., corn, wheat, soybean), distinct growth stages (e.g., vegetative, flowering, senescence), and characteristic disease or stress manifestations (e.g., nitrogen deficiency, fungal infection, water stress). The "view category specific quality assessment" neural network is a recurrent convolutional neural network (R-CNN) specifically designed to process spectral-spatial information, trained on a vast library of labeled hyperspectral signatures linked to ground-truthed plant physiological states and disease severity scores. The quality assessment value quantifies specific parameters such as normalized difference vegetation index (NDVI), chlorophyll content, water stress index, or pathogen severity. This output directly informs variable-rate application maps for precision fertilizer, pesticide, or irrigation systems, optimizing resource use and crop yield.
• Mermaid Diagram:

  graph TD
      A[Drone/Satellite HSI Sensor] --> B(Agricultural Field);
      B -- Hyperspectral Data Cube --> C{HSI Preprocessing & Radiometric Correction};
      C --> D[Hyperspectral Image (Spectral-Spatial)];
      D --> E{R-CNN (Crop Type/Growth Stage/Disease Specific)};
      E -- Crop Health Metrics (NDVI, Chlorophyll, Disease Severity) --> F[Precision Farming Management Platform];
      F --> G[Variable-Rate Application System];
  

4. Integration with Emerging Tech

Derivative 1.8: AI-driven Optimization with IoT-enabled Feedback Loop and Blockchain-verified Training Data

• Enabling Description: This system for echocardiographic image analysis incorporates real-time AI-driven optimization, IoT sensing, and blockchain for data integrity. "Signals representing a first at least one echocardiographic image" (Claim 1a) are acquired by an IoT-enabled ultrasound transducer (e.g., Clarius C3 HD3) equipped with integrated inertial measurement units (IMUs), force sensors for probe pressure, and patient position trackers. The "view category specific quality assessment value" (Claim 1c) is determined by an on-device neural network and is simultaneously fed into a proprietary generative adversarial network (GAN)-based AI optimization engine. This engine, running on a local GPU, generates real-time, context-aware recommendations for optimal transducer adjustments (e.g., angulation, rotation, depth, gain, frequency) which are presented to the sonographer via an augmented reality (AR) overlay on the ultrasound display. The training dataset for the neural networks (as per Claim 20) is immutably stored and managed on a permissioned blockchain ledger (e.g., Hyperledger Fabric). Expert quality assessment values and associated image metadata are cryptographically signed by certified echocardiographers and time-stamped as transactions on the ledger, ensuring data provenance, preventing tampering, and providing an auditable history for regulatory compliance. Furthermore, the AI optimization suggestions, along with the sonographer's acceptance or rejection of these suggestions, are also logged as transactions on the blockchain to continuously refine the GAN's recommendation model and improve its efficacy.
• Mermaid Diagram:

  graph TD
      A[IoT-enabled Transducer (IMU, Force Sensors)] --> B(Echocardiographic Image + Sensor Data);
      B --> C{On-Device View Category NN};
      B --> D{On-Device Quality Assessment NN};
      C & D --> E(Quality Assessment Value);
      E --> F{AI Optimization Engine (GAN) - Real-time Recs};
      F -- AR Overlay Display --> G[Sonographer];
      G -- Accepted/Rejected Adjustments --> H(Blockchain Ledger);
      H -- Verified Training Data --> I[Neural Network Trainer (Claim 20)];
      I --> C & D;
  

5. The "Inverse" or Failure Mode

Derivative 1.9: Low-Power, Limited-Functionality "Screening Mode" with Safe-Failure Protocols

• Enabling Description: This variation implements a "low-power screening mode" within the echocardiographic image analysis system, designed for rapid, preliminary cardiac assessment in austere, remote, or resource-limited environments. In this mode, the "at least one processor" (Claim 1) is a highly power-optimized ARM Cortex-M microcontroller (e.g., STM32L series) operating at a significantly reduced clock frequency. Non-essential peripherals and high-resolution imaging capabilities are disabled. The "view category specific quality assessment" neural network (Claim 1c) is replaced by a highly compressed and quantized neural network model (e.g., an 8-bit integer quantized MobileNetV3 architecture) with a drastically reduced parameter count, specifically optimized for binary classification: "adequate for basic screening" vs. "requires expert review/higher fidelity imaging." This quantization reduces computational and memory demands, extending battery life significantly (e.g., 24+ hours on a single charge). If the image quality determined by this simplified model falls below a pre-defined critical threshold for "adequate for basic screening," the system initiates a "safe-failure" protocol: it automatically saves the last 'N' (e.g., 5-10) low-resolution images and their associated metadata to a tamper-proof, non-volatile memory (e.g., secure eMMC). Concurrently, it displays a prominent "RE-SCAN REQUIRED: PROCEED TO ADVANCED IMAGING" message on a low-power monochromatic e-paper display and transmits a low-bandwidth alert (e.g., via LoRaWAN or NB-IoT) to a central telehealth monitoring station, providing an anonymized error code and location without patient-identifying data. The system automatically enters a diagnostic deep-sleep state if battery levels drop below a critical operational threshold, preventing data corruption or incomplete assessments.
• Mermaid Diagram:

  graph TD
      A[Low-Power Transducer] --> B(Reduced Resolution Image Stream);
      B --> C{ARM Cortex-M Microcontroller};
      C --> D{Quantized MobileNetV3 (Binary Quality Assessment)};
      D -- "Adequate / Requires Re-scan" --> E[Monochromatic E-paper Display];
      E -- Low Quality Trigger --> F{Safe-Failure Protocol};
      F -- Save N Low-Res Images --> G(Tamper-Proof eMMC);
      F -- LoRaWAN/NB-IoT Alert --> H[Central Telehealth Monitoring Station];
      C -- Critical Battery Level --> I(Diagnostic Deep-Sleep State);
  

---

Derivative Variations for Claim 20 (Computer-implemented system for training neural networks)

Core Claim 20: A computer-implemented system with at least one processor configured to receive echocardiographic training images (with view categories), receive expert quality assessment values (view-category-specific), and train neural networks using this data to determine network parameters, with at least a portion of each network associated with a view category.

1. Material & Component Substitution

Derivative 2.1: Federated Learning Framework with Homomorphic Encryption for Expert Feedback

• Enabling Description: This variation describes a computer-implemented system for training neural networks that utilizes a federated learning paradigm, enabling collaborative model training across multiple healthcare institutions while preserving strict data privacy. Instead of receiving "signals representing expert quality assessment values" (Claim 20b) directly at a central server, local clinical workstations (edge devices) at each institution perform initial, decentralized training steps. Each edge device contains powerful GPUs (e.g., NVIDIA H100) and processes its local "plurality of echocardiographic training images" and "expert quality assessment values." To ensure patient privacy and expert confidentiality, these expert quality assessment values and corresponding local model updates (e.g., gradients or weights) are homomorphically encrypted using a fully homomorphic encryption (FHE) library based on the Brakerski/Fan-Vercauteren (BFV) scheme before transmission. A central "trainer processor" (Claim 20) then performs encrypted aggregation of these homomorphically encrypted model updates from numerous participating institutions without ever decrypting the raw expert assessments or patient images. The "neural networks" are trained iteratively on these encrypted aggregates, generating updated "sets of neural network parameters" that are robust and generalized across a diverse, global clinical dataset while adhering to stringent privacy regulations (e.g., HIPAA, GDPR).
• Mermaid Diagram:

  graph TD
      subgraph Local Clinical Workstation 1 (Edge)
          A[Training Images 1 + Expert QAV 1] --> B{FHE Module (BFV Scheme)};
          B --> C(Local Model Training);
          C -- Encrypted Model Update --> D(Secure Channel);
      end
      subgraph Local Clinical Workstation N (Edge)
          E[Training Images N + Expert QAV N] --> F{FHE Module (BFV Scheme)};
          F --> G(Local Model Training);
          G -- Encrypted Model Update --> H(Secure Channel);
      end
      D & H --> I{Central Trainer Processor (Encrypted Aggregation)};
      I -- Iterative Encrypted Training --> J[Globally Optimized FHE-trained NN Parameters];
      J --> K[Secure Model Distribution];
  

Derivative 2.2: Quantum-Annealing-Accelerated Neural Network Training with Superconducting Qubits

• Enabling Description: This computer-implemented system for training neural networks leverages quantum annealing for accelerated optimization of neural network parameters. The "training the neural networks" (Claim 20c) involves framing the optimization of specific "sets of neural network parameters" (e.g., weights and biases of the shared convolutional layers or fully connected layers) as a Quadratic Unconstrained Binary Optimization (QUBO) problem. A D-Wave Advantage quantum annealer, utilizing superconducting flux qubits, is employed to find the ground state (optimal solution) or near-ground states of this QUBO problem. The "expert quality assessment values" (Claim 20b) are integrated into the cost function of the QUBO problem, where lower energy states directly correlate with improved model performance against expert labels. The classical "trainer processor" (Claim 20) manages the preprocessing of "echocardiographic training images" (Claim 20a) and expert labels, converts relevant neural network sub-problems into QUBO instances for the quantum annealer, orchestrates the annealing process, and then interprets the quantum annealing results to iteratively refine the neural network parameters in conjunction with classical backpropagation for other layers. This hybrid approach aims to mitigate issues like local minima and potentially accelerate convergence for complex, high-dimensional parameter spaces.
• Mermaid Diagram:

  graph TD
      A[Training Images + Expert QAV] --> B{Classical Preprocessing Unit};
      B -- Formulate QUBO Problem --> C{D-Wave Advantage Quantum Annealer};
      C -- Annealing Results (Parameter Candidates) --> D{Classical Post-processing / Backpropagation Unit};
      D -- Iteratively Refined NN Parameters --> E[Hybrid NN Model Storage];
      E --> F[Neural Network Deployment];
  

2. Operational Parameter Expansion

Derivative 2.3: Massive-Scale Distributed Training for Global Echocardiography Datasets (Exascale Computing)

• Enabling Description: This system is designed for training neural networks on an unprecedented scale, capable of processing petabytes of "echocardiographic training images" sourced from a global consortium of medical institutions. The "at least one processor" (Claim 20) represents an exascale computing environment, comprising millions of CPU cores and hundreds of thousands of specialized AI accelerators (e.g., GPUs, TPUs) interconnected by a high-bandwidth, low-latency InfiniBand EDR or custom optical network fabric. The "plurality of echocardiographic training images" (Claim 20a) could be in the order of billions, necessitating advanced data sharding, dynamic load balancing, and fault-tolerant distributed training algorithms (e.g., Asynchronous Stochastic Gradient Descent with parameter servers, communication-avoiding gradient compression techniques). The "expert quality assessment values" (Claim 20b) are generated through a meticulously standardized, multi-national, consensus-based labeling protocol involving numerous expert echocardiographers, ensuring consistency across diverse data sources. The training process employs customized PyTorch Distributed or JAX/XLA frameworks to optimize for exascale parallelism, allowing the determination of "sets of neural network parameters" for highly robust and generalizable models that account for significant variability in patient demographics, imaging device manufacturers, and acquisition protocols.
• Mermaid Diagram:

  graph TD
      subgraph Global Data Sources
          A[Hospital A (Images/QAVs)] --> D;
          B[Clinic B (Images/QAVs)] --> D;
          C[Research Lab C (Images/QAVs)] --> D;
      end
      D[Petabytes of Standardized Training Data] --> E{Exascale Computing Cluster (Millions Cores/Accelerators)};
      E -- Distributed Training (PyTorch/JAX) --> F[Globally Optimized, Generalizable NN Parameters];
      F --> G[Central Model Repository (Version Control)];
  

Derivative 2.4: Ultra-Low Latency, On-Device Retraining for Adaptive Patient-Specific Models

• Enabling Description: This system focuses on enabling ultra-low latency, continuous on-device retraining of neural networks for patient-specific echocardiographic image analysis. The "at least one processor" (Claim 20) is an integrated edge AI chip (e.g., Google Edge TPU, NVIDIA Jetson Orin Nano) directly embedded within a portable ultrasound device. "Echocardiographic training images" (Claim 20a) are local scans from the current patient, and "expert quality assessment values" (Claim 20b) are provided by the clinician in near real-time (e.g., via a touch interface) immediately after image acquisition. The training process, "train the neural networks" (Claim 20c), utilizes incremental learning combined with federated transfer learning or meta-learning techniques (e.g., MAML). This allows the pre-trained global model on the device to rapidly adapt its "view category specific portions" to the unique anatomical characteristics, physiological variations, or specific echogenicity of an individual patient, or to learn the subtle nuances of transducer manipulation by the current operator. The retraining cycle, from expert input to updated model parameters, completes within milliseconds to a few seconds, providing real-time personalized model improvements without requiring cloud connectivity. This ensures that the quality assessment is continuously optimized for the specific context of the current patient and operator.
• Mermaid Diagram:

  graph TD
      subgraph Portable Ultrasound Device
          A[Live Echocardiographic Image] --> B(On-Device Inference Model);
          C[Clinician Expert QAV (Real-time Input)] --> D(Edge AI Chip - Adaptive Retraining Module);
          B & C --> D;
          D -- Incremental Model Update (ms-s latency) --> B;
          B --> E[Real-time Patient-Specific Quality Feedback];
      end
  

3. Cross-Domain Application

Derivative 2.5: Training Neural Networks for Satellite Imagery Classification in Environmental Monitoring

• Enabling Description: The computer-implemented system for training neural networks is adapted for environmental monitoring through satellite imagery. The "plurality of echocardiographic training images" (Claim 20a) is replaced by multi-temporal, multi-spectral, and hyper-spectral satellite image data (e.g., from Sentinel-2, Landsat-8/9, MODIS) covering diverse geographic regions globally. The "plurality of predetermined echocardiographic image view categories" (Claim 20a) are analogous to environmental land cover and land use classes, such as "deforested area," "healthy forest canopy," "urban impervious surface," "water body with algal bloom," "wetland ecosystem," or "agricultural land with specific crop type." "Expert quality assessment values" (Claim 20b) are provided by environmental scientists, remote sensing specialists, or validated ground-truthing campaigns, labeling image segments based on ecological health indices, biomass estimations, or identified environmental anomalies. The "at least one processor" (Claim 20) trains specialized deep learning architectures (e.g., U-Net, DeepLabV3+ with spatial-temporal attention mechanisms) to determine "sets of neural network parameters" for accurate pixel-wise classification and semantic segmentation of these environmental categories. Portions of the neural networks are specifically tailored for different sensor modalities (e.g., optical vs. SAR), atmospheric conditions, or specific biomes.
• Mermaid Diagram:

  graph TD
      A[Multi-Temporal Satellite Images (Sentinel, Landsat)] --> B(Environmental Scientist Expert Labels);
      B -- Land Cover/Use Categories (e.g., Forest, Urban, Water) --> C{Neural Network Trainer (DeepLabV3+/U-Net)};
      C -- Train Modality/Biome-Specific NN Portions --> D[Environmental Monitoring NN Parameters];
      D --> E[Global Land Cover Atlas & Change Detection System];
  

Derivative 2.6: Training Neural Networks for Defect Detection in Semiconductor Manufacturing

• Enabling Description: This computer-implemented system for training neural networks is repurposed for automated defect detection in semiconductor manufacturing. The "plurality of echocardiographic training images" (Claim 20a) are replaced by high-resolution images acquired from various in-line inspection tools, including brightfield optical microscopes (e.g., KLA-Tencor), scanning electron microscopes (SEM), and atomic force microscopes (AFM) performing wafer-level metrology. These images capture critical features at different fabrication stages (e.g., after lithography, etch, deposition, chemical-mechanical planarization). The "plurality of predetermined echocardiographic image view categories" (Claim 20a) now corresponds to specific manufacturing layers or inspection steps (e.g., "polysilicon gate layer," "metal interconnect layer 3," "shallow trench isolation"), each having a distinct pattern and expected defect types. "Expert quality assessment values" (Claim 20b) are derived from experienced semiconductor engineers, process technicians, or high-accuracy automated defect classification (ADC) systems, labeling defects based on type (e.g., particle, scratch, bridge, void, pattern collapse), size, and location, with severity scores impacting device yield. The "at least one processor" trains specialized convolutional autoencoders or Vision Transformers (ViT) to determine "sets of neural network parameters" for robust detection and classification of microscopic defects, leading to real-time process control adjustments and significant yield improvements.
• Mermaid Diagram:

  graph TD
      A[Microscope/SEM/AFM Wafer Images] --> B(Semiconductor Engineer Expert Labels);
      B -- Fabrication Layer Categories (e.g., Lithography, Interconnect) --> C{Neural Network Trainer (Conv Autoencoder/ViT)};
      C -- Train Layer-Specific NN Portions --> D[Defect Detection NN Parameters];
      D --> E[Semiconductor Fabrication Process Control System];
  

Derivative 2.7: Training Neural Networks for Material Science Characterization (Microstructure Analysis)

• Enabling Description: The training system is adapted for material science research and industrial quality control, specifically for automated characterization of material microstructures. The "plurality of echocardiographic training images" (Claim 20a) are replaced by high-magnification images obtained from advanced microscopy techniques such as electron backscatter diffraction (EBSD) for crystallographic orientation, scanning transmission electron microscopy (STEM) for atomic-scale features, or X-ray microtomography (XMT) for 3D volumetric analysis of various material samples (e.g., superalloys, advanced ceramics, polymer composites). The "plurality of predetermined echocardiographic image view categories" (Claim 20a) corresponds to distinct microstructural features or phases, such as "ferrite grains," "austenite twins," "martensite laths," "grain boundary networks," "precipitate phases," or "void distributions." "Expert quality assessment values" (Claim 20b) are provided by metallurgists, material scientists, or image analysis specialists, annotating and quantifying critical microstructural parameters (e.g., grain size, phase fraction, defect density, aspect ratio) from these images, often correlated with mechanical properties or performance. The "at least one processor" trains specialized graph convolutional networks (GCN) or deep residual networks (ResNet) to determine "sets of neural network parameters" for automated identification, segmentation, and quantification of these microstructural features, accelerating materials discovery, process optimization, and failure analysis.
• Mermaid Diagram:

  graph TD
      A[EBSD/STEM/XMT Material Images] --> B(Metallurgist/Scientist Expert Labels);
      B -- Microstructural Feature Categories (e.g., Grain Boundary, Precipitate Phase) --> C{Neural Network Trainer (GCN/ResNet)};
      C -- Train Feature-Specific NN Portions --> D[Material Characterization NN Parameters];
      D --> E[Automated Materials Design & Quality Control Platform];
  

4. Integration with Emerging Tech

Derivative 2.8: Decentralized Training with Confidential Computing and AI Explainability for Regulatory Compliance

• Enabling Description: This computer-implemented system for training neural networks is designed for decentralized collaborative training across multiple hospitals or research centers, prioritizing data privacy and regulatory compliance. Each participating institution utilizes a secure "confidential computing enclave" (e.g., Intel SGX or AMD SEV technology) for local processing of its "plurality of echocardiographic training images" (Claim 20a) and "expert quality assessment values" (Claim 20b). Within this enclave, local model training occurs, ensuring that raw, sensitive patient data and expert annotations are never exposed outside the trusted execution environment. Only anonymized and aggregated model updates (e.g., gradient vectors or encrypted parameter deltas) are then shared with a central "trainer processor" (Claim 20). Furthermore, the training pipeline integrates AI Explainability (XAI) techniques (e.g., LIME, SHAP, integrated gradients) during the parameter determination process. This generates human-understandable justifications and feature importance maps for the "sets of neural network parameters," demonstrating why the model makes certain quality assessments. These XAI outputs, along with all data contributions and model updates, are immutably recorded as transactions on a distributed ledger technology (DLT) (e.g., Ethereum blockchain or a private consortium chain). This DLT provides a transparent and auditable record for regulatory bodies (e.g., FDA, EMA) to verify model integrity, data provenance, and ethical AI development in a highly regulated medical domain.
• Mermaid Diagram:

  graph TD
      subgraph Decentralized Training Node A
          A[Training Images A + Expert QAV A] --> B{Confidential Compute Enclave A};
          B -- Anonymized Model Update A + XAI Log A --> D(DLT Ledger);
      end
      subgraph Decentralized Training Node B
          C[Training Images B + Expert QAV B] --> E{Confidential Compute Enclave B};
          E -- Anonymized Model Update B + XAI Log B --> D;
      end
      D --> F{Central Aggregator / Trainer Processor};
      F -- Consolidated Training --> G[Explainable & Auditable NN Parameters];
      G --> H[Regulatory Compliance Platform (DLT Verification)];
  

5. The "Inverse" or Failure Mode

Derivative 2.9: Adversarial Training for Robustness Against Intentional Image Degradation (Cybersecurity Focus)

• Enabling Description: This variation of the computer-implemented system for training neural networks explicitly focuses on developing robustness against intentional image degradation or adversarial attacks. The "training the neural networks" (Claim 20c) incorporates an adversarial training loop involving a generative adversarial network (GAN). The "generator" component of the GAN acts as an "adversary," learning to produce subtle, imperceptible perturbations (adversarial examples) to the "echocardiographic training images" (Claim 20a). These perturbations are designed to mislead the "image quality assessment neural network" (acting as the "discriminator" in this GAN setup), causing it to output incorrect "quality assessment values." The "trainer processor" (Claim 20) then iteratively trains the "discriminator" network using a mix of clean "echocardiographic training images" and these synthetically generated adversarial examples. The objective is to force the discriminator to become highly robust and accurately determine "view category specific quality assessment values" even when presented with inputs that have been subtly manipulated. This enhanced training regimen ensures that the deployed image analysis system can withstand malicious attempts to compromise the integrity of echocardiographic data and maintain reliable diagnostic accuracy in the face of cybersecurity threats.
• Mermaid Diagram:

  graph TD
      subgraph Adversarial Training Loop
          A[Clean Echocardiographic Images + Expert QAV] --> B{Generator (Adversary)};
          B -- Adversarial Perturbations --> C[Adversarial Examples];
          C --> D{Image Quality Assessment NN (Discriminator)};
          A --> D;
          D -- Quality Assessment / Adversarial Loss --> E{Trainer Processor};
          E -- Update Discriminator & Generator --> D;
      end
      D -- Robust & Adversarial-Resistant NN Parameters --> F[Secure Deployment System (Cyberhardened)];
  

Derivative 2.10: Progressive Degradation Training for Low-Resolution / Compressed Image Inputs

• Enabling Description: This computer-implemented system for training neural networks is specifically designed to create models that maintain high accuracy in "view category specific quality assessment" even when presented with severely degraded or aggressively compressed "echocardiographic training images." The training process (Claim 20c) involves a progressive degradation module that applies various forms of artificial degradation to the original high-fidelity "echocardiographic training images" (Claim 20a). These degradations include, but are not limited to, JPEG/MPEG compression with varying quality factors (e.g., 5% to 80%), Gaussian noise injection with different signal-to-noise ratios (SNRs), resolution downsampling (e.g., from 512x512 to 64x64 pixels), and simulated packet loss or missing pixel data. For each synthetically degraded image, the corresponding "expert quality assessment values" (Claim 20b) are either derived from the original high-fidelity assessment or re-evaluated by experts to reflect the expected quality under that specific degradation. The "at least one processor" then trains the "neural networks" on this diverse dataset of progressively degraded images. This specialized training ensures that the deployed models can provide meaningful and reliable "quality assessment values" even when operating in scenarios with bandwidth constraints (e.g., telemedicine), limited computational resources (e.g., mobile ultrasound), or challenging acquisition conditions.
• Mermaid Diagram:

  graph TD
      A[High-Fidelity Echocardiographic Images] --> B{Progressive Degradation Module};
      B -- Degraded Image Versions (Compressed, Noisy, Low-Res) --> C[Expert QAV Re-assessment/Confirmation];
      C --> D{Neural Network Trainer};
      D -- Train on Degraded & Labeled Data --> E[Robust NN for Degraded Inputs];
      E --> F[Telemedicine / Mobile Ultrasound Deployment (Bandwidth-Limited)];
  

---

Combination Prior Art Scenarios

Here are at least three scenarios where US Patent 11129591 can be combined with existing open-source standards to establish prior art, demonstrating obviousness for future incremental improvements.

1. Combination with DICOM (Digital Imaging and Communications in Medicine) Standard for Interoperable Quality Assessment Reporting.

   • Description: The computer-implemented system described in US11129591 (specifically Claim 1 and its dependent claims for image analysis) is integrated with the DICOM standard. The "signals representing a first at least one echocardiographic image" (Claim 1a) are received as DICOM Part 10-compliant image files (e.g., US Multi-frame Image Storage SOP Class), ensuring seamless interoperability within existing medical imaging infrastructure (PACS). The "view category" (Claim 1b) is determined either by parsing existing DICOM attributes (e.g., (0018,5100) Patient Orientation, (0008,0008) Image Type, (0020,0062) Referenced Image Sequence pointing to view classification definitions) or by automatically generating a new DICOM attribute using a private tag within a standard group. The "first quality assessment value" (Claim 1c) is then generated by the system and encapsulated within a DICOM Structured Report (SR) object (DICOM Part 16, using IOD (0008,0016) ContentCreatorRole with a standard template such as TID 5200: Echocardiography Report or a custom template if necessary for the quality metric). This SR object is then associated with the original echocardiographic image series via standard DICOM object relationships and stored in a Picture Archiving and Communication System (PACS) or transmitted to an Electronic Health Record (EHR) system, enabling the quality assessment to be readily accessible and machine-readable across healthcare systems.
   • Open-Source Standard: DICOM (Digital Imaging and Communications in Medicine) Standard (e.g., using DCMTK or pydicom open-source libraries).
   • Relevance: DICOM is the fundamental and pervasive standard for medical imaging data exchange and storage. The integration of image analysis results, particularly quality assessments for clinical utility, into DICOM-compliant structures is a logical and obvious extension for any developer working within medical imaging informatics. The patent itself mentions PACS.

2. Combination with OpenCV (Open Source Computer Vision Library) and Keras/TensorFlow for Neural Network Implementation and Image Preprocessing.

   • Description: The systems and methods described in US11129591 (specifically Claim 1 for image analysis and Claim 20 for neural network training) extensively rely on neural networks, including convolutional layers, pooling layers, and LSTM layers. The "at least one processor" (Claim 1, 20) utilizes OpenCV (Open Source Computer Vision Library) for essential image preprocessing steps, such as filtering (e.g., cv2.medianBlur, cv2.bilateralFilter for speckle reduction), amplification (e.g., cv2.normalize), scan-conversion (e.g., geometric transformations using cv2.warpAffine), and resizing (e.g., cv2.resize). The "neural networks" (Claim 20c) themselves, including their architecture definition (e.g., Conv2D, MaxPooling2D, LSTM layers), the determination of "sets of neural network parameters," and the training process using a "mean absolute error loss function" and "stochastic gradient-based optimization algorithm" (as detailed in the patent), are explicitly implemented using the open-source deep learning framework Keras with a TensorFlow backend. The parameters (weights and biases) determined during training are stored in standard formats like HDF5 (Hierarchical Data Format 5), which is native to Keras for model persistence.
   • Open-Source Standard: OpenCV (Open Source Computer Vision Library) and Keras/TensorFlow.
   • Relevance: The patent explicitly references the use of the "Keras deep learning library with TensorFlowTM backend" for training and testing models. This direct mention, combined with the ubiquitous nature of OpenCV for image processing in any computer vision application, makes the combination of these open-source tools with the described inventive concepts an obvious engineering choice for anyone skilled in the art.

3. Combination with FHIR (Fast Healthcare Interoperability Resources) for Clinical Decision Support System Integration.

   • Description: The computer-implemented system for facilitating echocardiographic image analysis (Claim 1) produces "quality assessment values" that are intended to represent "suitability for quantified clinical measurement of anatomical features" and to "assist in diagnosing a medical condition or a characteristic of the heart". To enhance the utility and interoperability of these assessment values within the broader healthcare ecosystem, the system transmits or integrates this information using the FHIR standard. Specifically, the "first quality assessment value" (Claim 1c) and associated contextual data (e.g., patient identifier, study identifier, imaging event timestamp, the determined view category, specific metrics contributing to the quality score) are encapsulated within standard FHIR Resources. For instance, the overall quality assessment could be represented as a DiagnosticReport Resource, while individual component scores (e.g., LV, LA, MV scores, centering, depth, gain assessments as mentioned in the patent) could be represented as Observation Resources, linked back to the ImagingStudy Resource (which references the original DICOM images). This allows the quality assessment data to be consumed by other FHIR-compliant clinical systems, such as Electronic Health Records (EHRs), Clinical Decision Support (CDS) systems, or research databases, enabling automated workflows, improved data integrity checks, and more informed clinical decisions. The communication adheres to FHIR RESTful APIs and uses standard JSON or XML formats.
   • Open-Source Standard: FHIR (Fast Healthcare Interoperability Resources).
   • Relevance: FHIR is rapidly becoming the industry standard for healthcare data exchange, particularly for clinical data intended for real-time use and integration. The generation of a clinically relevant quality assessment value by US11129591 logically leads to its integration into FHIR resources to ensure maximum interoperability and utility, a step that would be considered obvious by a person skilled in the art of health informatics.