US 10458895

I'll search for authoritative information on US 10458895, including any 2026 CAFC docket activity.

Let me refine the searches to target USPTO and litigation records specifically.

The initial searches returned mostly noise. Let me try targeted queries for the claims and for any litigation.

Bottom line

US 10,458,895 B2 is a Case Western Reserve University patent on a radiomics/machine‑learning method for predicting whether an NSCLC patient will respond to pemetrexed chemotherapy, using shape and texture ("radiomic") features extracted from baseline CT images of both the tumor and a ring of tissue immediately surrounding it (the "peritumoral" region).

I found no CAFC 2026 docket activity for this patent, and I could not retrieve the verbatim issued claim text. Details and caveats below.

Bibliographic summary

Abstract (as published)

"Methods, apparatus, and other embodiments predict response to pemetrexed based chemotherapy. One example apparatus includes an image acquisition circuit that acquires a radiological image of a region of tissue demonstrating NSCLC that includes a region of interest (ROI) defining a tumoral volume, a peritumoral volume definition circuit that defines a peritumoral volume based on the boundary of the ROI and a distance, a feature extraction circuit that extracts a set of discriminative tumoral features from the tumoral volume, and a set of discriminative peritumoral features from the peritumoral volume, and a classification circuit that classifies the ROI as a responder or a non-responder using a machine learning classifier based, at least in part, on the set of discriminative tumoral features and the set of discriminative peritumoral features."

Independent claims — plain-language overview

⚠️ Uncertainty flag: The document I worked from (Google Patents full text) is truncated before the claims section, and my searches did not surface the verbatim claim set. The descriptions below are therefore reconstructed from the abstract and the specification's method/apparatus embodiments, not from issued claim language. Treat them as a functional summary, not a claim chart.

Based on the abstract and the described embodiments, the independent claims appear to fall into these groups:

1. Independent apparatus claim (mirrors the abstract). A computing apparatus comprising:

• an image acquisition circuit that acquires a pre‑pemetrexed‑treatment radiological image (CT) of NSCLC tissue containing an annotated ROI defining a tumoral volume with a boundary;
• a peritumoral volume definition circuit that defines a peritumoral volume from the ROI boundary plus a distance (specification examples: 7 pixels, 2.5 mm, or 5 mm; also described as a heuristic or tumor‑property‑dependent dilation);
• a feature extraction circuit that extracts a set of discriminative tumoral features (shape and texture) and a set of discriminative peritumoral features (texture); and
• a classification circuit that classifies the ROI/patient as a responder or non‑responder using a machine‑learning classifier (LDA, QDA, KNN, random forest, or SVM, per the specification), optionally also returning a probability and optionally selecting features/classifier based on CT slice thickness.

1. Independent method claim — personalized treatment planning (specification "Method 100"). Steps of: accessing a pre‑chemotherapy CT image with an annotated tumoral ROI (110) → defining a peritumoral volume from the boundary (120) → extracting discriminative tumoral features including shape and texture (130) → extracting discriminative peritumoral textural features (140) → providing both feature sets to a machine‑learning classifier (150) → receiving a responder/non‑responder classification (160) → generating a personalized treatment plan (e.g., a pemetrexed schedule or dosage) based on the classification (170).

2. Independent method claim — predicting chemotherapy response (specification "Method 900"). Steps of: accessing a pre‑treatment image containing a segmented tumoral volume and segmented peritumoral volume (910) → extracting a discriminative feature set that includes at least two tumoral features and at least one peritumoral feature (e.g., mean of intratumoral sum average, minimum of intratumoral Law, sphericity, mean of peritumoral Law) (920) → providing them to a classifier trained and tested on disjoint independent training/validation sets of responder and non‑responder pre‑treatment images (930) → receiving a responder/non‑responder classification (940).

3. Possible independent computer‑readable‑storage‑device claim. The specification states that the methods (100, 200, 300, 900) may be stored as computer‑executable instructions on a non‑transitory computer‑readable storage device, which is a common claim format in this family; whether such a claim actually issued independently I cannot confirm.

Notable technical content that likely appears in dependent claims or the specification: extraction of ~1,542 radiomic features; Haralick, Laws, Laws‑Laplacian, Gabor, LBP/DLBP, HOG/DHOG, CoLiAGe, and 3D shape/sphericity feature families; feature selection via T‑test, Wilcoxon, MRMR, and feed‑forward ranking; slice‑thickness‑dependent feature/classifier selection; and reported performance of ≥0.76 AUC (reported classifier accuracies ranged ~52%–78% across KNN, LDA, QDA, SVM‑linear/RBF/polynomial, and RF).

Litigation / CAFC 2026 docket status

• No appeal or CAFC docket entry for US 10,458,895 was found. My searches of the open web, the CAFC‑related results surfaced by search, and USPTO‑related document hits did not return any appellate proceeding, IPR, or district‑court case involving this patent number. If a 2026 Federal Circuit appeal exists, my searches did not surface it, and I cannot confirm it.
• Explicit non‑match (do not conflate): A USPTO PTAB petition (Champion Power Equipment, dated Aug. 20, 2025) repeatedly refers to an asserted "'895 Patent" in a portable‑generator/power‑equipment patent family (alongside U.S. Patents 10,221,780; 10,393,034; 10,697,398; 11,143,120; 11,143,145; 11,492,985; 11,530,654; 11,840,970). That '895 is a different patent from US 10,458,895 — the technology, parties, and family are unrelated to this medical‑imaging patent. I am flagging it only so it is not mistaken for litigation activity against US 10,458,895.
• Other search hits (Apple/Taction haptics involving U.S. 10,659,885; Contour IP v. GoPro; Cerence v. Samsung; Spatial Genomics v. 10X Genomics) involve visually similar but numerically distinct patent numbers and are not US 10,458,895.

Confidence and uncertainty statement

• High confidence: title, assignee, inventors, application/filing/issue dates, provisional priority, abstract, CPC classes, and the technical substance of the disclosure (all taken from the authoritative Google Patents full text supplied).
• Medium confidence: the mapping of independent claims described above — it is reconstructed from the abstract and the numbered embodiments in the specification, not from the issued claims.
• Low / no confidence: the existence and exact wording of a computer‑readable‑storage‑device independent claim; the exact number and ordering of independent claims; and any legal‑status or docket facts. Google Patents expressly characterizes its legal‑status and expiration entries as assumptions, and I found no authoritative docket confirming or denying 2026 CAFC activity.

If you need the verbatim claims, I'd recommend pulling the granted‑patent PDF from USPTO PatentCenter (application 15/612,467) or the USPTO Patent Public Search full‑text view — I could not retrieve that text through the searches available to me here.

I'll search for litigation involving US patent 10458895.

Let me search more targeted sources.

Let me check litigation-specific databases directly.

Litigation search results for US Patent No. 10,458,895

Bottom line: I found no litigation — no district court suits, no CAFC appeals, no ITC proceedings, and no PTAB/IPR challenges — involving US Patent No. 10,458,895 ("Predicting response to pemetrexed chemotherapy in non-small cell lung cancer (NSCLC) with baseline computed tomography (CT) shape and texture features").

Searches performed

I ran targeted searches for the literal identifiers 10458895, 10,458,895, and US10458895B2 against patent-litigation-oriented sources (Unified Patents litigation portal, CourtListener/RECAP, Docket Alarm, general web). None of the results returned an infringement action, declaratory judgment action, ITC investigation, or AIA post-grant proceeding naming this patent.

Important caveat on search quality: the identifier searches returned either (a) the patent's own Google Patents page, or (b) false hits on different numeric strings — e.g., U.S. Application Ser. No. 09/512,895, U.S. Patent 5,648,985, and an unrelated Delaware case (Orca Security Ltd. v. Wiz, Inc., 1:23-cv-00758-JLH). Per the strict instruction not to auto-correct or conflate identifiers, I have excluded all of these; subsections of larger serial numbers are not the same as the patent number 10,458,895.

Verified facts about the patent (from the authoritative patent text)

What I can and cannot say

• What I can say with reasonable confidence: There is no publicly docketed litigation or PTAB proceeding involving this patent. It is a university-owned, federally funded diagnostic/radiomics patent that was issued in 2019, and I found no record of enforcement activity. CWRU newsroom coverage describes it purely as an issued patent and technology-licensing item, not as the subject of a suit.
• What I cannot say with high confidence: I could not directly render the Unified Patents litigation case-list page or run a live CourtListener/PACER docket query for this exact number within my tool budget. My negative finding is therefore based on the absence of any indexed litigation record in search results, not on a positive "no cases" report pulled from a docket database.

Recommended verification before relying on this for any legal purpose: run the exact string 10,458,895 (and 10458895) through (i) CourtListener's RECAP docket search, (ii) Docket Alarm / Lex Machina / Docket Navigator, (iii) the USPTO Patent Center "Litigation" and PTAB APIs, and (iv) the ITC EDIS system — filtering out the unrelated serial-number and patent-number false hits identified above.

If no case is found there either, the accurate statement is: no litigation is known as of the date of this search (April 26, 2026).

PTAB Proceedings Analysis — US Patent 10,458,895

Patent: US 10,458,895 B2 ("Predicting response to pemetrexed chemotherapy in NSCLC with baseline CT shape and texture features")
Patent Owner: Case Western Reserve University
Inventors: Madabhushi, Velcheti, Orooji, Rakshit, Alilou, Beig
App. No.: 15/612,467 | Filed: 2017-06-02 | Priority: 2016-06-06 | Granted: 2019-10-29 | Adjusted expiration: 2038-01-09
Source of record: USPTO Open Data Portal structured "PTAB proceedings on file" block (canonical list)

Proceedings overview

Total AIA trial proceedings on file: 0 (zero). Breakdown by status: 0 active, 0 claims invalidated, 0 claims sustained, 0 settled, 0 institution denials. The ODP structured block for this patent returns no AIA trial proceedings, and independent web research (Google Patents, PTAB/PTACTS filings, general PTAB reporting) surfaced no IPR, PGR, or CBM involving US 10,458,895. The defensive posture this creates for a defendant is "no PTAB precedent to lean on — and no estoppel to inherit." Every claim of the '895 patent stands untested at the Board, so the patent is neither "hardened" by surviving challenges nor "softened" by having claims canceled. A defendant being asserted against today is starting from a blank slate: no claim of this patent has ever been construed or judged by an APJ panel, and no petitioner has ever borne (and lost) the burden of proving unpatentability.

Important false-positive flag: A search hit for "IPR2023-01344" and "'895 patent" refers to U.S. Patent 9,575,895 (Daedalus Prime LLC v. Mercedes-Benz), a different patent with a coincidentally similar number. It has no relationship to US 10,458,895. Do not cite it.

Second flag: the ODP legal-status line "Active, expires 2038-01-09" is a Google Patents/ODP status annotation and reflects patent term and maintenance-fee status — not any validity proceeding. It is not evidence of PTAB activity.

Proceedings

No proceedings to report. The required per-proceeding format (petitioner/patent owner, judge panel, grounds, institution decision, FWD, settlement, appeal, defensive value) cannot be populated because no AIA trial proceeding exists on this docket. Per the task constraint — "Don't fabricate. If the FWD didn't address a claim, don't claim it did" — no proceeding entries, panels, FWD dispositions, or appeal dockets are invented here.

Strategic summary

Claim status: all claims UNTESTED. Because no AIA trial has ever been instituted, no claim of US 10,458,895 has been canceled, confirmed, or even construed by the Board. Claims 1 et seq. remain exactly as granted on 2019-10-29. There is no narrowed, surviving claim set to work with, and no canceled claim set to attack. Any assertion letter citing these claims must be met on the merits of the claims themselves, not by pointing to a PTAB outcome.

Estoppel landscape: a clean field — which cuts both ways. § 315(e)(2) estoppel bars a petitioner (and its privies) from raising in district court any ground it raised or reasonably could have raised in an IPR that reached an FWD. Here, there has been no FWD, so no party is estopped as to this patent and no defendant inherits an estoppel-defense argument against the patent owner (there is no prior petitioner whose invalidity theories the patent owner has already beaten). Conversely, there is no free ride: a defendant contemplating an IPR must build its own grounds from patents and printed publications (§ 311(b) — IPR grounds are limited to §§ 102/103 over patents and printed publications; § 112 and public-use/on-sale art are off the table in an IPR). The prior-art space for radiomic/pathomic feature extraction and machine-learning classification of CT tumor and peritumoral regions is well developed in the literature, and Google Patents' own citation record for this family shows the "Prior art keywords" tags of features / discriminative / peritumoral / tumoral / volume — a dense field that a diligent petitioner would need to mine. Note also the family dimension: the '895 patent sits in the Madabhushi/CWRU radiomics portfolio, and sibling claims in related continuations may be the more efficient IPR target depending on which claims are actually asserted. Constraint observed: I have not verified the specific sibling filings or their claim sets, so do not rely on that point without independent confirmation.

Pattern signals: none yet, and that is the signal. There is no repeat petitioner, no defensive aggregator (no Unified Patents, RPX, or similar) in the chain, and no evidence of the patent owner aggressively appealing Board outcomes — because there has been no Board outcome to appeal. This is characteristic of a university-held, industry-non-practicing diagnostic patent that has not yet been the subject of an assertion campaign sufficient to draw IPR fire. Well-asserted patents eventually attract IPRs; the absence here suggests either (a) limited litigation-driven assertion activity to date, or (b) assertions that have settled before a petition ripened. I cannot confirm which from the sources searched.

Recommended next steps

If you are a defendant and there is no PTAB activity (the situation here): say so plainly in your risk memo — no AIA trial proceeding on US 10,458,895 exists on the ODP record as of this analysis. That means:

• No FWD to link or quote. There is no disposition to cite, because none exists. Any representation that a claim is canceled would be fabrication.
• PGR is time-barred. The patent is PGR-eligible (effective filing date after 2013-03-16), but the § 321(c) nine-month post-grant window closed around 2020-07-29 (nine months after the 2019-10-29 grant). Do not plan on PGR.
• CBM is unavailable. The patent is a medical-diagnostics/radiomics method, not a "covered business method" tied to financial products or services, and the CBM program sunset for new petitions on 2020-09-16 in any event.
• IPR is the only live AIA vehicle, grounded solely in patents and printed publications under §§ 102/103.
• Watch the § 315(b) clock. If your client has been served with a complaint alleging infringement of this patent, an IPR petition must be filed within one year of service or you are barred. Calendar this immediately.
• Trial-stage milestones if you file: the Board's institution decision is due within three months of the patent owner's preliminary response (or the last date such a response could be filed, 37 C.F.R. § 42.107(b)); if instituted, the FWD is due within one year of institution (§ 316(a)(11)) absent good cause. Track the proceeding on USPTO PTAB E2E/PTACTS and docket any appeal at CourtListener / the Federal Circuit.

Sources consulted:

• USPTO ODP structured "PTAB proceedings on file" block (canonical, returns zero proceedings)
• Google Patents, US10458895B2: https://patents.google.com/patent/US10458895B2/en
• USPTO PTAB E2E / PTACTS public information portal: https://ptacts.uspto.gov/ptacts/ (no filings for this patent number)
• Federal Circuit opinions via CourtListener: https://www.courtlistener.com/ (no appeal from this patent found)

Confidence: High that no AIA trial proceeding existed as of the most recent ODP ingest and no PTAB/CAFC activity is reachable via public web sources. Moderate-to-high that no recently-filed, not-yet-indexed petition exists — a very recent petition (within the ODP ingest lag) cannot be ruled out by web search alone. If precision matters for a filing or litigation deadline, confirm directly against PTAB E2E by patent number before relying on this.

Verification note on method

Per the operating rules, patent and proceeding identifiers are interpreted literally and not auto-corrected. The near-miss hit at IPR2023-01344 concerns U.S. Patent 9,575,895 (petitioner Mercedes-Benz entities; patent owner Daedalus Prime LLC) and is expressly not treated as a proceeding on US 10,458,895. No proceeding number, panel, FWD, or settlement has been invented to fill the empty record. Where a fact could not be verified (sibling family filings, whether any assertion has settled pre-petition), that is stated as an open item rather than asserted.

I'll research the assignment chain for US 10,458,895, starting with the USPTO assignment records and cross-referencing corporate/litigation sources.

Good — I have the primary bibliographic data. Now let me pin down the specific assignment reel/frame and correspondent for this patent, plus check for any post-issuance transfers.

Important data-availability caveat

I could not retrieve the individual reel/frame numbers or the recorded correspondent for this patent's assignment entry. Google Patents' legal-events table and the USPTO bulk-index search results I could reach expose the fact of the assignment and the parties, but not the reel/frame or correspondence block. Everything below that is stated as fact comes from the authoritative Google Patents record supplied earlier in this analysis plus the fetched full text; anything inferred is flagged as such. I have not fabricated reel/frame numbers.

Inventors

Unusual-pattern note (soft): There is no documented mass inventor departure within 12 months of the 2016-06-06 priority date — the normal fire-sale precursor is absent. What is observable is ordinary academic mobility over later years: Madabhushi has since moved to Emory University and is associated with a radiomics spin-out ("Picture Health"), and Velcheti has since moved institutions. These are post-filing career moves, not evidence of a coordinated exit, and I could not confirm exact departure dates or that any inventor assigned away from CWRU. Treat the mobility observation as context only.

Original assignee

Case Western Reserve University (CWRU), a private research university at 10900 Euclid Avenue, Cleveland, OH 44106 — named as applicant (71) and assignee (73) on the face of the patent.

• Primary line of business: Higher education and academic research; it monetizes IP through its technology-transfer office rather than by selling products.
• Product embodying the claims: No. The claims cover a computer-implemented radiomics/machine-learning method (and corresponding apparatus) for predicting pemetrexed response; CWRU does not commercialize a clinical product embodying them. Any exploitation is by license to third parties (potential licensee class: imaging-AI/CADx vendors, pathology/radiology software firms).
• Current status: Operating (university, active). No bankruptcy, acquisition, or dissolution. CWRU remains a prolific patent filer (its portfolio includes the related radiomics family, e.g. the NSCLC/peritumoral radiomics work visible in the assignee listings retrieved).

No public-company assignor/assignee is in this chain, so there are no 10-K/8-K disclosures to reconcile (CWRU files Form 990, not SEC annual reports).

Assignment timeline

⚠️ This patent's Assignment Center record is, as far as I could reach it, a single-link chain. The only recorded assignment is the initial inventor→university assignment.

• Executed ~2017 (exact execution date not retrieved) / recorded 2017-07-14 — Reel not retrieved (reel/frame not exposed by the sources I could reach)
  • Conveyance: Assignment of Assignors' Interest ("ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS)" per Google Patents)
  • Assignor: Mehdi Alilou; Mahdi Orooji; Sagar Rakshit; Vamsidhar Velcheti; Niha Beig; Anant Madabhushi (all six named inventors)
  • Assignee: Case Western Reserve University (Cleveland, OH)
  • Correspondent: Not confirmed for this reel. Based on a separate, unrelated CWRU recording I could retrieve (US app 15/344,959, Reel 040243 / Frame 0471, recorded 2016-11-07, submitter Craig W. Hayden, correspondent firm Tarolli, Sundheim, Covell & Tummino LLP, 1300 East Ninth Street, Suite 1700, Cleveland, OH 44114; email mkalchoff@tarolli.com), CWRU's routine assignment correspondent of record is the Tarolli firm. This is a pattern inference for CWRU filings generally, not a confirmed correspondent for US 10,458,895. I flag it because a single CWRU-typical correspondent on a single university-recorded assignment is a normal in-house/outside-counsel arrangement, not an NPE tell.
  • Context: Standard academic invention assignment — all inventors conveying to their university/clinical-collaborator institution at filing. Not a fire-sale, reorg, securitization, or transfer-to-asserter.

No post-issuance assignment record (no transfer to an LLC, IP holding entity, asserter, or defensive aggregator) was found. If such a record exists, my sources did not surface it. The absence of any post-issuance link ordinarily means the original assignee still owns the patent, which for a university patent is the expected result.

No Assignment Center multi-link chain to diagram beyond the founding assignment.

Timeline diagram

timeline
    title Ownership of US 10458895
    2016 : Priority date 2016-06-06
         : Inventors assign rights to CWRU
    2017 : Application filed 2017-06-02
         : Assignment recorded 2017-07-14
    2019 : Patent issued 2019-10-29
         : No post-issuance transfers recorded

(The "inventors assign rights to CWRU" line reflects that the recording occurred in 2017; the execution date was not separately retrieved, so it is shown conservatively in the 2016–2017 window.)

NPE / troll-pattern signals

Because the chain contains only the founding university assignment and no downstream transfer, nearly every troll signal is absent by construction. Marking each per the required scale:

1. Shell-entity transfer — Not present. The only recorded assignee is Case Western Reserve University (reel not retrieved; recorded 2017-07-14). No "IP / Holdings / Licensing / Ventures" LLC, registered-agent service address, or single-purpose Delaware/Texas entity appears anywhere in the chain.
2. Known asserter in the chain — Not present. CWRU does not appear on the RPX / Unified Patents / Marathon / Acacia / Intellectual Ventures / Wi-LAN / Conversant-type NPE rosters, and no such entity appears as a prior or current assignee. (No litigation involving this patent number was surfaced.)
3. Repeat correspondent across the chain — Not present. With only one link, there is no recurrence test to run. The CWRU-typical correspondent (Tarolli, Sundheim, Covell & Tummino LLP — Reel 040243/0471 on a different patent) is unconfirmed for this reel and, even if confirmed, is a mainstream Cleveland IP firm doing university prosecution work, not an NPE recording attorney.
4. Cascading transfers — Not present. No chained LLC transfers in any window; zero post-issuance links.
5. Pre-litigation transfer — Not present. No infringement suit naming this patent was found, so there is no triggering suit and therefore no suspiciously timed pre-suit assignment.
6. Bankruptcy fire-sale — Not present. CWRU is an operating university; no Chapter 7/11 event. (Aside: don't conflate this with the unrelated Champion Power Equipment PTAB matter that refers to a different "'895 Patent" in a portable-generator family — that is not US 10,458,895.)
7. Privateering — Not present. No operating-company-to-NPE transfer and no SEC coverage revealing an assertion arrangement.
8. Defensive aggregator (anti-NPE) — Not present. Chain does not terminate at RPX, AST, LOT, Unified Patents, or OIN; it terminates at the university.

Verdict

Insufficient data (only the original assignment).

Justification: the only recorded ownership event I could reach is the 2017-07-14 inventor→Case Western Reserve University assignment of assignors' interest (Google Patents legal events; reel/frame not exposed by the sources available). There is no post-issuance transfer to any LLC, NPE, or aggregator, and no litigation naming US 10,458,895 was found. The presence of a single founding university assignment with no downstream links is itself the finding: CWRU appears to remain the owner of record, and none of the eight NPE signals is present.

Verification links: Assignment Center search page — https://assignmentcenter.uspto.gov/ ; legacy public index — https://assignment.uspto.gov/patent/index.html (search "10458895" or application "15/612,467" to retrieve the reel/frame and correspondent that I could not extract here).

Residual uncertainty: I could not confirm the exact reel/frame, execution date, or correspondent, and I could not positively exclude a very recent (2025–2026) recordation. Pulling the granted-patent PDF or the Assignment Center entry driven by patent number 10,458,895 / application 15,612,467 would close that gap.

I'll research the prior art citations for US 10,458,895, focusing on USPTO records and the patent's "References Cited" sections.

Let me dig deeper into the specific cited references and USPTO records.

Let me try to retrieve the patent's citation tables directly from patent databases.

Bottom line

I was unable to retrieve the actual "References Cited" / "Patent Citations" list printed on the face of US 10,458,895 from the sources available to me. My searches of Google Patents, FreePatentsOnline, patents‑review, and general web did not return the examiner's cited‑reference table (the "U.S. Patent Documents," "Foreign Patent Documents," and "Other Publications" rows) for this specific patent. The authoritative full‑text you supplied earlier is truncated before the claims and the references‑cited section, so it does not contain that list either.

Per the operating rule that I must not fabricate identifiers or references, I will not invent a citation list or assign § 102 anticipation to references I have not actually confirmed are cited by (or applicable to) US 10,458,895.

Below I separate (a) what I could verify, (b) the citation‑family documents I did find and can legitimately describe, and (c) the explicit gaps.

What I could verify

Any reference qualifying as § 102(a)/(b)/(e) prior art must be dated before that priority/filing date. This is the filter I would apply once the actual list is in hand — and it is also why the documents below that post‑date 2016 are not § 102 art here.

Documents I found — and why they are NOT § 102 prior art to this patent

A. Forward citations (patents that cite US 10,458,895 — these are later documents)

These appeared in my searches as documents citing US 10,458,895. Because they post‑date the 2016 priority date, they cannot anticipate US 10,458,895:

• US 2020/0258218 A1 — "Method based on deep neural network to extract appearance and geometry features for pulmonary textures classification." Its citation table lists US10458895B2 as a cited document. This is a forward citation.
• US 2021/0110541 A1 — "Combination of radiomic and pathomic features in the prediction of prognoses for tumors" (Vaidya, Bera; CWRU). Its "Citations (10)" table lists US 2017/0351939 A1 (the pre‑grant publication of the instant patent). Again a forward citation.

These do not qualify as § 102 prior art and should not be presented as anticipating claims of US 10,458,895.

B. Same‑inventor / same‑family documents (also not § 102 art)

• US 2019/0156954 A1 — "Methods and apparatus for predicting benefit from immunotherapy using tumoral and peritumoral radiomic features" (Madabhushi et al., CWRU). Same inventors, closely related subject matter (tumoral + peritumoral radiomic features, peritumoral dilation of 2.5 mm/5 mm/7 pixels). This is a sibling/continuation‑family‑type disclosure. Sharing inventors and overlapping priority, it is not available as § 102 prior art against US 10,458,895, though it is highly relevant to the family and to any obviousness‑type comparison.
• Other Madabhushi/CWRU patents surfaced (e.g., US 10,078,895; US 11,464,473; US 11,335,0901 etc. as listed on the Med‑Emory lab page and PatentLeaderboard) are related‑field art but are not confirmed citations on the face of US 10,458,895.

I am explicitly flagging that none of section A or B is a confirmed "reference cited" against US 10,458,895, and none is § 102 prior art given the dates.

What I could NOT retrieve (the actual task deliverable)

The task asks for each reference cited in US 10,458,895 with:

• full citation,
• publication/filing date,
• brief description, and
• the claim(s) it potentially anticipates under § 102.

I could not obtain the cited‑reference list, so I cannot supply any of those four fields for any reference without fabricating them. I will not do that.

Recommended authoritative sources to obtain the real list (in priority order)

1. USPTO Patent Public Search (ppubs.uspto.gov) — full‑text view of US 10,458,895; the "References Cited" section and the "Patent History / File Wrapper" tab hold the examiner's citations (Form PTO‑892).
2. USPTO PatentCenter — application 15/612,467: the "Documents & Transactions" tab contains the examiner's search notes and any Form 892 listing cited art.
3. Google Patents "Citations" panel for US10458895B2 (the "Patent Citations" + "Non‑Patent Citations" sub‑tables) — normally located below the description; the page did not render this section into my search results.
4. FreePatentsOnline / Espacenet — "Also published as / Citing documents / Cited documents" tabs.

Honest assessment / confidence

• High confidence: the bibliographic facts, priority/filing dates, and the analysis that forward citations (2020, 2021) and same‑inventor family members are not § 102 prior art to this patent.
• No confidence / cannot state: the identity, dates, and § 102 relevance of any reference actually cited on the face of US 10,458,895. I have no verified citation list and will not guess.
• Contradiction flagged with the prior analysis: the earlier "Patent summary" also noted it could not retrieve the verbatim claims or full citation set from the truncated Google Patents text. That remains true here — nothing in my subsequent searches closed that gap. There is no contradiction between the two sections on this point; both flag the same missing data.

If you can paste the "References Cited" section (or let me run a new session targeting the USPTO PatentCenter document list for application 15/612,467), I can complete the four‑field, claim‑by‑claim § 102 mapping you asked for.

I'll research the prior-art landscape before constructing the § 103 analysis, since the task directs me to use prior-art results and I need to ground any combination in real references.

Good — I have the core radiomics prior art. Let me pull a few more pre-2016 references to ground the specific claim elements (peritumoral features, chemotherapy-response prediction, classifier selection).

Obviousness Analysis — US 10,458,895 B2 under 35 U.S.C. § 103

⚠️ Threshold caveat / contradiction flag (must be read first). The task instructs me to "use the results from the Prior Art section of this page," but no formal "Prior Art" section was supplied in my materials. The only prior-art artifacts in the authoritative patent text are the Google Patents front-matter entries Prior art keywords: features, discriminative, peritumoral, tumoral, volume and the "Background" section. I therefore treat the externally verified pre‑priority radiomics literature I located as the prior-art basis, and I flag every reference whose date or content I could not directly confirm. Second, the earlier-generated sections state that the verbatim issued claim text could not be retrieved; this analysis therefore operates on the reconstructed independent claims (apparatus; "Method 100"; "Method 900"), not on literal claim language. Treat the claim mapping as representative, not as a claim chart.

1. Governing framework and effective prior-art date

• The application (US 15/612,467) was filed 2017‑06‑02, claiming benefit of provisional 62/345,960 (2016‑06‑06). It is post‑AIA, so §§ 102/103 as amended apply.
• The presumptive critical date is 2016‑06‑06. A reference qualifies as § 102(a)(1) art if it was publicly available before that date. If any given claim is not supported by the provisional, its effective date slides to 2017‑06‑02 (which only adds available art). Either way, the references below antedate the critical date.
• Graham/KSR framework: scope and content of the prior art → differences → PHOSITA level → secondary considerations. Under KSR, a combination is obvious where (i) the elements were known, (ii) there was a known problem for which the combination was an obvious solution, and (iii) the results were predictable; "a finite number of identified, predictable solutions" and "ordinary creativity/routine experimentation" defeat non-obviousness.

PHOSITA definition. A person with an M.S. or Ph.D. in biomedical engineering, medical physics, or computer science, or a thoracic radiologist/oncologist with several years of quantitative-imaging experience, familiar with (a) CT image acquisition and segmentation (e.g., 3D Slicer), (b) radiomic feature families (Haralick/GLCM, Laws, Laws‑Laplacian, Gabor, LBP, HOG, 3D shape/sphericity), (c) standard feature-selection and classifier toolkits, and (d) NSCLC chemotherapy response assessment. This is a routine level of skill in the 2014–2016 radiomics field.

2. The candidate prior-art references

Not prior art (flag): Khorrami et al., Radiology: Artificial Intelligence 1(2):e180012 (published 2019‑03‑20) — "Combination of Peri- and Intratumoral Radiomic Features on Baseline CT Scans Predicts Response to Chemotherapy in Lung Adenocarcinoma." This paper reports 1,542 features, a 26‑responder/27‑non‑responder training set, and QDA AUC 0.82/0.77 — i.e., it is effectively the same work, by the same inventors (Madabhushi, Velcheti, Alilou et al.), published after the priority date. It is therefore not § 102 art, but it is the best available evidence of what the claims in fact cover and should be used to scope the analysis — not to invalidate.

3. Element-by-element mapping to the reconstructed independent claims

4. The two central combinations

Combination A — "Radiomics platform + response prediction + ML toolbox" (strongest)

Primary references: Aerts 2014 + Coroller 2016; secondary: Parmar 2015a, Parmar 2015b; tertiary: Lambin 2012, Fedorov 2012.

Motivation, stated with KSR rationales:

1. Same field, same problem. Aerts 2014 and Coroller 2016 both address CT radiomics in lung cancer and share authors/institutions and a common research program — combining them is combining references "directed to the same problem."
2. Aerts supplies every feature family recited (shape, texture, intensity) from pre‑treatment CT, on a 1,019‑patient cohort with independent validation. Coroller supplies the response endpoint — the only meaningful difference from Aerts' prognostic endpoint. A PHOSITA seeking a predictive (not merely prognostic) radiomic biomarker in NSCLC would, with reasonable expectation of success, substitute Coroller's response label for Aerts' survival label. Predictable result.
3. Parmar 2015a supplies the entire ML layer verbatim: the same five classifiers named in the specification (LDA, QDA, KNN, RF, SVM), the same feature-selection families (Wilcoxon, mRMR), the same training/validation partitioning, and the same AUC evaluation. KSR: "a combination of familiar elements according to known methods" yielding "predictable results."
4. Parmar 2015b supplies consensus clustering — the specification's own methodology — by name.
5. Fedorov 2012 supplies the annotation tool (3D Slicer), which the specification concedes it "may" use.

Net: every element is disclosed; the only act is applying a known radiomics pipeline to a known response endpoint, using a known classifier toolbox. Under KSR, this is a textbook obvious combination.

Combination B — "Peritumoral features" combination

Primary: Aerts 2014 (or Coroller 2016) in view of the tumor-microenvironment biology literature and standard morphological dilation practice.

Motivation:

• The analytical step — "define a region around the ROI by dilating its boundary by a distance" — is elementary mathematical morphology, and the specification itself frames the distance as heuristically chosen (7 px, 2.5 mm, 5 mm are all recited as alternatives, i.e., a finite set of predictable options).
• The biological rationale for interrogating peri-tumoral tissue (peritumoral lymphatic microvessel density, peritumoral inflammatory infiltrate, vascular invasion/neovascularization) was well documented before 2016 and is explicitly recited in the specification's own background logic. A PHOSITA with that rationale and an existing texture extractor would have been motivated to apply the extractor to the dilated ring.
• Result predictable: once peritumoral HU/texture data is extracted, whether it is discriminative is determined by the same feature-selection machinery (Parmar 2015a/b), not by inventive insight.

Conclusion for peritumoral element: obvious as (i) known morphological dilation + (ii) known texture extraction + (iii) a known biological motivation to look outside the tumor.

5. Why the specific feature sets and numbers do not rescue the claims

A common non-obviousness argument would be that the claims recite particular features ("mean of intratumoral sum average," "minimum of intratumoral Law," "sphericity," "mean of peritumoral Law") and particular performance (≥0.76 AUC). Under KSR this is weak:

• Those features are the output of the very feature-selection methods recited in the claim (T‑test, Wilcoxon, mRMR, feed‑forward) — all of which are disclosed in Parmar 2015a. Where "the claimed invention is the product of ordinary experimentation within a disclosed method," it is obvious.
• The feature families (Gabor, Laws, Laws‑Laplacian) and the distribution statistics (mean, median, variance, skewness, kurtosis) are enumerated as known in the specification itself and in Aerts 2014.
• The ~1,542-feature library is a quantity, not a technical advance; choosing a library size is routine.
• Reported accuracy (~52 %–78 % across classifiers) and "≥0.76 AUC" are results, not structure; a result achieved by routine application of known tools does not confer patentability. Indeed, Coroller 2016 already reports AUC ≈ 0.61–0.63 for response, establishing that the field expected — and obtained — discriminative radiomic response features.

6. Where the § 103 case is weak / genuinely contestable (candor required)

1. The pemetrexed-specific hook. None of the pre‑2016 references I verified discloses pemetrexed-specific response prediction. Coroller (neoadjuvant chemoradiation) and Cook (chemoradiotherapy) are combination/radiation regimens. A patentee would argue that "predicting a specific drug's benefit" is a non-obvious selection. Counter: the specification's own data show a generic radiomic response signature, and pemetrexed is simply the standard-of-care first-line cytotoxic regimen (per the patent's own Background), so choosing it is a design choice among a finite set of standard regimens.
2. Date tightness for Coroller 2016. Epub 2016‑04‑13 vs. critical date 2016‑06‑06 — it qualifies, but only if the provisional supports the claim at issue. If any claim relies on matter added in the 2017 non‑provisional, the critical date moves to 2017‑06‑02 and Coroller remains art; the analysis is robust, but the weight of Coroller changes.
3. No verified pre‑2016 lung peritumoral‑radiomics reference. I could not, within my tool budget, confirm a specific published pre‑2016 paper extracting radiomic texture features from a peritumoral ring in lung CT. (The well-known peritumoral radiomics works — Dou 2018, Braman 2017, Khorrami 2019 — are all post-date.) The peritumoral element therefore rests on biological motivation + standard morphology + Parmar 2015a/b machinery rather than a single on-point reference. This is the most attackable part of the prima facie case and would benefit from locating a pre‑June‑2016 disclosure.
4. "Generate a personalized treatment plan" (element G). This is arguably a mental/routine step and/or an abstract concept; it also raises § 101 questions (Mayo/Alice) that are outside a pure § 103 analysis but that a challenger would pair with § 103.
5. Secondary considerations. None apparent: the technology was in a crowded, fast-moving academic field; the specification itself admits "conventional approaches… provide no better results than guessing," which cuts against a long-felt-but-unmet-need narrative for the broad concept (though it supports a narrow one). No evidence of commercial success, copying, or industry praise was located.

7. Bottom line

• On the reconstructed independent claims (apparatus; Method 100; Method 900), a strong prima facie § 103 case exists. The best combination is Aerts 2014 + Coroller 2016 + Parmar 2015a (machine learning) + Parmar 2015b (consensus clustering), optionally with Lambin 2012/Fedorov 2012, and with the peritumoral element supplied by standard morphological dilation in view of pre‑2016 tumor-microenvironment biology.
• Every element is disclosed; the only "invention" is the application of a known radiomics pipeline to a known NSCLC response endpoint using a known classifier/feature-selection toolbox, with a predictable result — the classic KSR combination.
• The vulnerable elements are (i) the pemetrexed-specific limitation, (ii) the peritumoral-radiomics element (no verified on-point pre‑2016 reference), and (iii) the specific feature/performance recitations if the issued claims are as narrow as the Khorrami 2019 paper suggests.
• Highest-confidence statement: US 10,458,895's core concept — intra‑ and peritumoral CT radiomic texture/shape features fed to a machine-learning classifier to classify NSCLC therapy responders vs. non‑responders — was squarely within the prior art and the ordinary skill level as of June 2016.
• Lowest-confidence statement: whether the issued claims (which I could not read verbatim) are broad enough to be fully anticipated/obvious, or whether their specific feature sets and pemetrexed limitation carve out a non-obvious remainder. Resolving this requires the granted claim text from USPTO PatentCenter (application 15/612,467), which I could not retrieve.

Verification recommended before relying on this for any legal purpose: (1) pull verbatim claims from USPTO Patent Public Search/PatentCenter; (2) confirm the exact publication date and content of Cook 2013 and any 2014–2016 lung peritumoral-radiomics reference; (3) confirm the provisional 62/345,960 disclosure to lock the critical date.