Skip to content
Computational Pathology Reading Roadmap

Evaluation and clinical translation

A computational-pathology model is not “validated” once it performs well on a held-out set. Internal discrimination, external robustness, calibrated risk, pathologist assistance, workflow benefit, patient benefit, and safe deployment are different claims that require different studies.

Latest literature discovery audit: 2 October 2026. Existing paper reviews retain their individual source-check dates.

This chapter is a study-design and evidence route, not regulatory or medical advice. Use the applicable institutional, jurisdictional, ethical, laboratory, and device requirements for a real clinical project.

1. Start with an intended-use statement

Section titled “1. Start with an intended-use statement”

Complete this before choosing a metric:

user: pathologist | technician | researcher | tumour board | autonomous system
population: inclusion, exclusion, specimen, disease spectrum, sites
input: stain, scanner, slides/case, metadata available at prediction time
output: localization | diagnosis | grade | biomarker | risk | triage | QC
workflow_role: replace | assist | prioritize | rule-out | second read | research only
action: what changes after the output
comparator: unaided practice or current test
failure_policy: abstention, fallback, repeat test, human review

Without this contract, the same AUROC may be interpreted as a diagnostic device, research biomarker, or prioritization tool even though those uses have different costs and operating points.

2. Match the claim to the evidence

Section titled “2. Match the claim to the evidence”
Evidence stage Study design What it can support What it cannot establish alone
Technical verification unit tests, synthetic cases, pipeline replay software and transformation correctness clinical performance
Internal retrospective locked patient-level test from development sources performance on sampled source population new-site generalization
External retrospective independent institution/scanner/time period transport to that external setting live workflow benefit
Silent prospective model runs on incoming cases without influencing care real-time data compatibility and drift effect on users or patients
Reader/workflow study representative users, cases, crossover or randomized design change in accuracy, time, agreement, workload long-term patient benefit
Prospective interventional model changes workflow or decisions clinical/process outcomes under protocol universal deployment
Post-deployment monitoring continuous performance, safety, drift, incidents sustained local performance and emerging harms performance outside monitored use

Call a cohort “external” only when its patients and acquisition process are independent of development and potential pretraining overlap has been audited. For foundation models pretrained on public WSI collections, an evaluation cohort may be familiar even if the downstream head never saw it.

3. Design the split around biology and workflow

Section titled “3. Design the split around biology and workflow”

Split at the highest linked unit:

patient → case → specimen → block → slide → region → tile

Then audit:

  • repeated/re-cut slides and serial sections;
  • multiple stains or scanners for the same tissue;
  • matched primary/metastatic or longitudinal samples;
  • patients shared across public releases;
  • institutional batches and acquisition dates;
  • encoder pretraining and benchmark overlap; and
  • target labels derived after the intended prediction time.

Strong test axes

Section titled “Strong test axes”
Axis Question
Patient holdout does it generalize beyond the individual?
Site/scanner holdout does it survive acquisition and population shift?
Temporal holdout does it work on future practice and protocols?
Disease/subtype holdout is transfer broad or confined to represented phenotypes?
Specimen holdout does resection performance transfer to biopsy/frozen/cytology?
Prospective stream can the full system process real incoming cases and failures?

Nested cross-validation is useful for small cohorts, but it does not manufacture an external population. Keep all model selection—including thresholds, calibration, feature selection, stain references, and early stopping—inside training data.

4. Choose metrics from the decision

Section titled “4. Choose metrics from the decision”

Classification and screening

Section titled “Classification and screening”

Report class prevalence, confusion matrix, sensitivity, specificity, PPV, NPV, and confidence intervals at prespecified operating points. Add AUROC and precision–recall curves, but do not let a threshold-free summary replace the intended action.

For a rule-out system, estimate the fraction safely excluded at the required sensitivity and analyze every false negative. For triage, measure delay and queue effects, not only classification.

Localization and dense prediction

Section titled “Localization and dense prediction”

Use lesion/object matching, FROC or per-object sensitivity, instance/PQ metrics, boundary errors, and slide-level outcomes as appropriate. A high tile AUROC does not prove that all clinically relevant lesions are localized.

Ordinal grade and agreement

Section titled “Ordinal grade and agreement”

Use an ordinal confusion matrix, weighted agreement, and clinically meaningful under/over-grading categories. Compare against inter-reader variability with a consensus or adjudicated reference standard.

RACR-MIL is a useful ordinal case study: rank-aware attention and hybrid local/non-local WSI context target ordered squamous-cell-carcinoma grades across several anatomic sites. Preserve per-anatomy patient splits, ordinal versus nominal baselines, class prevalence, weighted agreement, confidence intervals, and the small pathologist-pilot denominator. The arXiv record reports Medical Image Analysis acceptance, but the final DOI/issue was not available at the cutoff; acceptance does not turn a reader pilot into clinical deployment evidence.

Risk and survival

Section titled “Risk and survival”

Report discrimination and calibration at relevant horizons, censoring-aware error, clinical baselines, decision curves where meaningful, and uncertainty. Do not choose risk-group cutoffs on the final test set.

Calibration and abstention

Section titled “Calibration and abstention”

Calibration is population- and prevalence-dependent. Report reliability by site and subgroup, recalibration policy, and coverage–risk curves for selective prediction. An abstaining model needs a workflow: who reviews the case, how often, and with what turnaround cost?

MVC-Bench adds a controlled medical-VLM calibration matrix spanning histopathology, fundus, and chest radiography. Report ECE together with MCE/ACE, binning, classwise reliability, accuracy, prompt/tuning conditions, and held-out shift. Its multi-class margin regularization is evidence for the tested classification protocols—not a guarantee for segmentation, report generation, VQA, or downstream decisions.

Generative and translation systems

Section titled “Generative and translation systems”

A synthetic pathology image needs a claim-matched evidence stack rather than one “realism” score:

Evidence layer Representative peer-reviewed reading Keep separate
Patch distribution Morphology-focused DDPM (2023) fidelity, coverage, genotype adherence, morphology, and memorization
Large-image composition DiffInfinite (2023) local seams, global mask/layout validity, pathologist discrimination, downstream utility, and formal privacy
Preparation translation F2FLDM (2025) target-domain appearance, source identity, case-aware distance, downstream classification, and hallucination
Conditional synthetic training Semantic and visual crop-guided diffusion (2025) map/crop adherence, encoder-dependent distance, synthetic-only task transfer, exemplar leakage, and reader validity
Text/prototype control UniPath (2026) visible semantics, prototype retrieval provenance, distribution metrics, expert preference, downstream augmentation, and overlap

Use patient-disjoint real test data for utility, retain the original image for every translation, report every feature encoder used by FID/FD-like metrics, and test copying or membership explicitly. A synthetic-only training result does not prove privacy; a pathologist’s real/synthetic confusion does not prove diagnostic safety.

Current claim-specific stress tests

Section titled “Current claim-specific stress tests”
Claim being audited Primary reading What must remain separate
fine-grained multiscale MLLM understanding PathView-Bench (2026, preprint) answer accuracy, visual localization, magnification/scale choice, and source overlap
slide- and patch-level OOD rejection ZIO (2026, preprint; code) prototype/support construction, threshold calibration, near versus far OOD, and retained in-distribution coverage
patient-level signal in a PFM atlas What Carries the Signal? (2026, preprint) patient, slide, tissue-composition, segmentation, and representation effects
dense transfer beyond classification PFMs as mitotic-figure detection encoders (2026, preprint; code) object matching, annotation policy, detector head, feature level, density, domain, and species shift
distributional cell-composition robustness Beyond Counts (2026, preprint) mean counts, full distributions, cell-model error, patient hierarchy, and the selected transport/margin definition
risk-controlled generated reports CorePath (2026, preprint) factual generation, conformal release/abstention, coverage, residual unsupported findings, and external workflow
survival lower-bound guarantees Censoring-aware conformal bounds (2026, preprint) discrimination, calibration, censoring assumptions, marginal/subgroup coverage, and clinical action
image-risk regions as molecular niches AI-guided spatial proteomics in TNBC (2026, preprint) retrospective risk prediction, spatial discovery in two patients, protein signature derivation, and outcome validation
expert veterinary/toxicologic visual knowledge VIPER (2026 preprint; code/data) question validity, ROI grounding, species/domain transfer, subgroup denominators, and the explicit patch-not-WSI boundary
native-language slide-to-report generation German prostate report VLM (2026 preprint) case-level pairing and split, attribute factuality, unsupported findings, linguistic coverage, and clinician adjudication

5. Expose shortcuts and subgroup failures

Section titled “5. Expose shortcuts and subgroup failures”

The study by Howard et al. (2021) demonstrates that submitting site can leave a detectable digital histology signature even after common color controls. Therefore:

  1. train a site/scanner predictor from the same representations;
  2. tabulate target prevalence by site and time;
  3. inspect nearest neighbours and attention regions with metadata;
  4. evaluate per site, scanner, demographic group, and specimen subtype;
  5. test color, compression, blur, tissue-content, and resolution perturbations; and
  6. repeat key results with site-stratified or leave-one-site-out designs.

Subgroup analysis needs adequate sample size and uncertainty. A non-significant difference is not proof of fairness. Predefine clinically relevant groups and document missing demographic data rather than inferring sensitive attributes from images.

6. Evaluate the complete system

Section titled “6. Evaluate the complete system”

An encoder benchmark does not evaluate a clinical pipeline. Lock and report:

  • slide reader, QC, tissue mask, and exclusion rules;
  • MPP, tile size, sampling, and stain intervention;
  • exact encoder architecture, weights, license, and preprocessing;
  • frozen, adapter, linear-probe, or full-fine-tuning policy;
  • aggregator, context budget, and slide/case pooling;
  • threshold and calibration model;
  • hardware, latency, memory, storage, and failed-case rate; and
  • human interface and fallback behavior.

For foundation-model comparisons, use identical splits and downstream protocols, multiple seeds, confidence intervals, a conventional supervised/ImageNet baseline, and compute-normalized as well as performance-focused analyses. One encoder can be strong for retrieval and weak for biomarker prediction; avoid a universal rank.

The broad comparison by Neidlinger et al. (published online 2025; volume 2026) is useful because it treats foundation encoders as feature extractors across controlled weakly supervised tasks. Read it alongside the specific model papers rather than copying one average into a leaderboard.

7. Landmark computational-pathology evidence

Section titled “7. Landmark computational-pathology evidence”

Read these for study design, not as proof that a current model is clinically ready.

Order Primary study Year / status Evaluation lesson
1 Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases (CAMELYON16) 2017 · JAMA challenge test set, lesion and slide metrics, and a pathologist comparison
2 Impact of Deep Learning Assistance on Histopathologic Review of Lymph Nodes 2018 · American Journal of Surgical Pathology assisted versus unassisted reader performance and review time
3 Clinical-Grade Computational Pathology Using Weakly Supervised Deep Learning 2019 · Nature Medicine scale, minimally curated data, held-out institutions, and a rule-out framing
4 Automated Deep-Learning System for Gleason Grading of Prostate Biopsies 2020 · Lancet Oncology consensus reference, observer study, and external material
5 AI Assistance Significantly Improves Gleason Grading 2021 · Modern Pathology pathologist–AI interaction and crossover-style assistance evaluation
6 Site-Specific Digital Histology Signatures 2021 · Nature Communications confounding and bias from acquisition site
7 Benchmarking Foundation Models as Feature Extractors for Weakly Supervised Computational Pathology 2025 · Nature Biomedical Engineering controlled encoder comparison and task-dependent rankings
8 Articulate Pro prostate-biopsy reporting study 2026 · npj Digital Medicine prospective multicenter observational NHS workflow evidence; not randomized and not patient-outcome evidence
9 Histology AI for recurrence and chemotherapy benefit 2026 · The Lancet Oncology multicenter retrospective development/validation using TAILORx and external cohorts; trial material does not make the AI intervention randomized
10 Histology Angioscore for anti-angiogenic response 2025 · Nature Communications retrospective validation with real-world and IMmotion150 material; preserve treatment-selection and prospective-evidence limits
11 Enterprise automated WSI QC deployment 2026 · Journal of Pathology Informatics complete-system operational evidence over 94,995 slides, with single-center and commercial-system limitations

For each, reconstruct the patient flow, reference standard, split, operating point, human comparator, confidence interval, and claim. Then list what further evidence would be needed for your intended use.

8. Reader and workflow studies

Section titled “8. Reader and workflow studies”

A reader study should define:

  • representative case spectrum, including difficult and negative cases;
  • reader expertise and number;
  • reference standard and adjudication;
  • randomized reading order, washout, and carryover controls;
  • unaided versus aided interface and training;
  • primary accuracy/agreement and time outcomes;
  • clustering by reader and case in the analysis;
  • handling of model abstention and interface failures; and
  • whether the model changes confidence, over-reliance, or error type.

The human–AI team is the intervention. Evaluating a heatmap in isolation misses alert fatigue, anchoring, automation bias, and workflow integration.

9. Reporting and risk-of-bias standards

Section titled “9. Reporting and risk-of-bias standards”

Use the guideline that matches the study; none is a substitute for sound design.

Resource Version Use it for
CLAIM 2024 update 2024 transparent medical-imaging AI development/evaluation reporting
TRIPOD+AI 2024 prediction-model development and evaluation reporting
PROBAST+AI 2025 model quality, risk of bias, and applicability assessment
STARD-AI 2025 diagnostic-accuracy studies of AI-based tests
DECIDE-AI 2022 early-stage live clinical evaluation of AI decision support
SPIRIT-AI 2020 protocols for trials of AI interventions
CONSORT-AI 2020 reports of clinical trials of AI interventions
FUTURE-AI 2025 lifecycle guidance for trustworthy, deployable healthcare AI
TRIPOD-LLM 2025 reporting studies using LLMs and VLMs; use beside, not instead of, task-specific imaging guidance
DPA clinical-laboratory recommendation 2026 separate scanner validation, algorithm validation, intended use, implementation, and clinical utility
CAP WSI validation guideline online 2021 / issue 2022 validation of the complete diagnostic WSI platform, not an AI model
IMDRF GMLP final principles FINAL:2025 international medical-device ML lifecycle principles
FDA predetermined change-control-plan guidance final, August 2025 U.S.-specific planned AI-device changes and post-deployment maintenance

Use EQUATOR to check for updates and study-specific extensions at submission time.

Do not call ISO/CD 24051-2 a published international standard: it remained a committee draft under development at the 2 September 2026 cutoff.

10. Reproducibility and release checklist

Section titled “10. Reproducibility and release checklist”

Cohort and labels

Section titled “Cohort and labels”
  • eligibility, exclusions, dates, sites, specimen types, and patient flow;
  • label source, timing, assay, annotators, adjudication, and missingness;
  • split at patient/case level with duplicate and overlap audit;
  • class/event prevalence and subgroup counts in every split.

Model and pipeline

Section titled “Model and pipeline”
  • complete WSI preprocessing and QC manifest;
  • exact model/checkpoint, code commit, dependencies, and seeds;
  • hyperparameter-selection boundary and number of experiments;
  • threshold and calibration fit without test leakage;
  • failed inputs and abstentions included in denominator.

Statistics

Section titled “Statistics”
  • prespecified primary metric tied to intended use;
  • confidence intervals with clustering/resampling at the patient level;
  • baseline and comparator on identical cases;
  • multiplicity policy for many tasks, genes, endpoints, or subgroups;
  • sensitivity analyses for site, MPP, stain, sampling, and missingness.

Transparency

Section titled “Transparency”
  • model card with intended use, prohibited use, limits, license, and version;
  • dataset and overlap/provenance statement;
  • code or sufficiently detailed reproducible protocol;
  • representative failure gallery without protected information;
  • privacy, ethics, conflicts, and funding disclosures.

11. A claim audit for every paper

Section titled “11. A claim audit for every paper”

For each paper, fill one row:

Claimed conclusion Unit/split Test domain Comparator Calibration Human/workflow evidence Supported?
example: “generalizes across hospitals” patient one external hospital same pipeline absent none only to the tested hospital; broader claim unsupported

Then ask:

  1. Did the evaluation population match the title and abstract?
  2. Could site, scanner, tissue amount, or label timing explain the result?
  3. Was every transformation fit without test information?
  4. Does the metric correspond to the clinical action?
  5. Were uncertainty and failed cases reported?
  6. Is attention or a saliency map being overstated as causal explanation?
  7. What is the next evidence stage?

12. Suggested capstone

Section titled “12. Suggested capstone”

Take one public slide-level task and build a locked evaluation package:

  1. immutable patient/site/temporal splits;
  2. WSI and feature manifests;
  3. conventional and foundation-encoder baselines under identical MIL;
  4. calibration and a prespecified operating point;
  5. site/scanner and subgroup stress tests;
  6. patient-level bootstrap confidence intervals;
  7. failure and abstention review; and
  8. a CLAIM/TRIPOD+AI-aligned report plus model card.

The capstone is complete only when another researcher can reconstruct which patients, slides, tiles, weights, thresholds, and decisions produced every reported result.

Risk-specific multi-teacher survival adapters

SurvMILKD combines frozen encoder features through trainable MIL adapters, teacher-space feature translation and soft ordinal risk distributions. It removes teacher branches at deployment while retaining one student feature pipeline.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
SurvMILKD: A Weakly Supervised Survival Analysis Framework for Multi-Teacher Knowledge Distillation using Pathology Foundation ModelsMayur Mallya, Ali Khajegili Mirabadi, Hossein Farahani, and Ali Bashashati2026 · ECCV · acceptedThe official ECCV 2026 poster and conference-hosted manuscript verify acceptance; a Springer version of record was not verified. The official repository was a code-coming-soon placeholder on 2026-09-12. All foundation encoders remain frozen: task-specific MIL adapters and translation modules learn, not the UNI backbone.Risk-aware multi-teacher distillation combines complementary slide-level survival signals while deploying only a student branch. SurvMILKD uses frozen pathology foundation encoders, trainable attention-MIL adapters, teacher-specific feature translators, and a four-bin representation of scalar risk. Its contribution is task-specific survival transfer rather than another general-purpose encoder pretraining recipe.Official ECCV 2026 Poster · Official repository — code pending

Reading question: How does distillation represent censored survival risk, and which teacher and feature costs remain in training versus inference?

Evidence boundary: Not every cohort benefits from more teachers; frozen encoders, external risk pooling and code-release status must remain explicit.

Taxonomy-conditioned diagnostic label generation

TaxoMIL generates coarse and fine diagnostic strings using a shared decoder, branch-specific prefixes and hierarchy-aware losses. Its target is a known label pair, so compare it with hierarchical MIL and exact-match diagnosis rather than free-text report generation.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
TaxoMIL: Taxonomy-Constrained Learning for Hierarchical Whole Slide Image AnalysisChaeyeon Lee, Khang Nguyen Quoc, Jinsol Song, Yosep Chong, Kwangil Yim, and Jin Tae Kwak2026 · ECCV · acceptedOfficial ECCV 2026 poster and conference-hosted manuscript verify acceptance; no Springer version of record was verified. Author spelling follows the paper rather than the abbreviated program entry. Output is coarse/fine diagnostic label text, not full clinical reports. GastWSI is private.TaxoMIL learns to generate a coherent pair of coarse and fine diagnostic labels from a whole slide. It combines MIL aggregation, two granularity-specific visual prefixes, a shared GPT-2 decoder with separate heads, and taxonomy-aware training losses. The key teaching idea is to structure both the label space and its alignment with visual evidence before judging hierarchical predictions.Official ECCV 2026 Poster · Official implementation · arXiv manuscript

Reading question: Does each prediction satisfy both granularity-specific correctness and the supplied parent–child taxonomy?

Evidence boundary: Taxonomy-aware training is not a formal guarantee for arbitrary generated strings; private cohorts and predefined labels limit transfer claims.

Cross-magnification transfer: utility, predictability and task benefit

After XMaG and LRMIL, test whether observed teacher targets help, whether native low-power features predict them, and whether MIL benefits from the predictions. Direct versus residual regression and native-feature retention are separate controls.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
What Makes High-Magnification Knowledge Transferable? A Study of Cross-Resolution Distillation in Whole-Slide ImagingZhiyuan Yang, Jiahao Cheng, and Mahdi S. Hosseini2026 · arXiv · preprintPublic arXiv v2 (2026-10-01) verified 2026-10-02. The authors state ICLR 2027 under review; no acceptance inferred. Only public source material was used. A diagnostic study of target utility, predictability and downstream benefit, not a new universally superior distillation architecture.Separate teacher usefulness, student predictability and downstream benefit. Observed 20× regional means can help slide prediction when supplied with native 5× features, yet a 5× predictor may fail to reconstruct them usefully. Lower reconstruction error does not consistently translate into better slide-level performance.Reviewed manuscript revision

Reading question: Can the proposed target beat a constant training-mean baseline and improve the final slide endpoint?

Evidence boundary: Observed 20× access is a diagnostic reference, not low-power-only deployment. Lower reconstruction MSE is not guaranteed downstream improvement; the study is an ICLR 2027 submission-stage preprint.

Selecting encoders before expensive MIL sweeps

SAM-Cluster and SAM-LP compare feature extractors using structure-aware regions and proxy evaluations before full MIL training. They provide a teacher-selection starting point, not a guaranteed downstream ranking.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
SAM-Cluster and SAM-LP: Structure-Aware Evaluation Metrics for Selecting WSI Feature Extractors Without MIL TrainingJuhyeon Kim, Paul Ssemakula, Chanjae Song, and Mun Yong Yi2026 · MICCAI · peer-reviewedOfficial MICCAI 2026 open-access proceedings version. Springer DOI and final page range were not listed at the audit date.SAM-Cluster and SAM-LP rank frozen WSI feature extractors before committing to full multiple-instance learning (MIL) training. SAM supplies structural regions, not tissue-class truth. The label-free score asks whether embeddings respect those regions; the supervised score learns which region prototypes support the slide label. The product is a candidate ranking for a particular cohort, not a new diagnostic classifier.Official accepted-version PDF · Official implementation and benchmark results

Reading question: Does a proxy ranking agree with the target dataset, endpoint and aggregation protocol?

Evidence boundary: Correlation is not identical rank or clinical validation. Keep SAM segmentation, extractor inference, label requirements and final MIL confirmation in the resource account.

Pathology concepts: distinguish representation from classifier reliance

Early pathology SAE work relates features to biological annotations and encoder depth. PATH-DISSECT adds depth-resolved selectivity checks and controlled interventions inside the frozen encoder, measured through a frozen linear probe. A represented tissue concept need not be used by every downstream head.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Learning biologically relevant features in a pathology foundation model using sparse autoencodersNhat Minh Le, Ciyue Shen, Neel Patel, Chintan Shah, Darpan Sanghavi, Blake Martin, Alfred Eng, Daniel Shenker, Harshith Padigela, Raymond Biju, Syed Ashar Javed, Jennifer Hipp, John Abel, Harsha Pokkalla, Sean Grullon, and Dinkar Juyal2024 · arXiv · preprintReview uses arXiv:2407.10785v3, including its supplement. A later NeurIPS 2025 Imageomics workshop paper has the title Evaluating the Utility of Sparse Autoencoders for Interpreting a Pathology Foundation Model; its inaccessible full PDF was not used to replace this version's claims or publication status.A sparse dictionary trained on frozen PLUTO embeddings exposes cell- and tissue-associated directions that are easier to inspect than individual dense coordinates. The study links selected SAE activations to estimated cell counts, compares layers and stains, and checks transfer to different cancer cohorts. Its main contribution is an empirical interpretability study, not a new foundation encoder or proof that every sparse feature is one biological concept.Versioned full manuscript · Authors' later Imageomics workshop publication
PATH-DISSECT: Interpreting Concept Representation and Classifier Reliance in Pathology Foundation Models with Sparse AutoencodersMayur Parvatikar, Xingjian Li, and Min Xu2026 · MICCAI MI4MedFM Workshop · acceptedOfficial MICCAI 2026 Satellite Events page supplies the 11-page camera-ready; Springer DOI and final pages were pending at inspection. This is MI4MedFM workshop work on tissue patches, not main-conference or patient-level WSI outcome validation. Causal claims concern interventions on a frozen encoder plus linear readout, not biological causation.PATH-DISSECT separates a concept being encoded from a classifier relying on it. It trains sparse autoencoders at multiple blocks of frozen pathology transformers, validates class-associated features, then edits those features inside the forward pass while measuring a frozen linear probe. Its intervention is on model activations, not a causal experiment on tissue biology.Official camera-ready PDF · Author implementation · Workshop OpenReview record

Reading question: Is a tissue concept encoded, and does the specified classifier actually depend on it?

Evidence boundary: Tile-level annotations and linear-probe interventions are not WSI outcome validation or causal tissue biology. Preserve patch sampling, patient/site splits, feature-selection scope, and error-preserving controls.

Sparse features for slide prototypes and weakly supervised localization

ProtoMIL uses SAE-derived patch prototypes in slide learning; SAE-1N selects a discriminative latent using slide supervision. These methods connect sparse features to WSI tasks, but localization quality and slide-level predictive performance can move in different directions.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Prototype-Based Multiple Instance Learning for Gigapixel Whole Slide Image ClassificationSusu Sun, Dominique van Midden, Geert Litjens, and Christian F. Baumgartner2025 · MICCAI · peer-reviewedOfficial MICCAI 2025 record and 11-page accepted manuscript checked. PANDA is evaluated as binary tumor classification in this study, not the original six-class ISUP grading task. Artifact concepts are masked before MIL retraining; this is not a universal AUC improvement.ProtoMIL discovers sparse visual concepts from frozen CONCH patch embeddings, then expresses a WSI class logit as a sum of weighted concept contributions. Pathologists can inspect maximally activating patches and disable unwanted concepts before retraining the slide classifier. The point is an inspectable decision decomposition, not a claim that attention alone explains a prediction.Official MICCAI open-access record · Official accepted manuscript · Author implementation · arXiv manuscript
Patch-level phenotype identification via weakly supervised neuron selection in sparse autoencoders for CLIP-derived pathology embeddingsKeita Tamura, Yao-zhong Zhang, Yohei Okubo, and Seiya Imoto2026 · Pacific Symposium on Biocomputing · peer-reviewedPSB 2026 volume 31, pages 708–721, with source supplement checked. Neuron discovery uses slide labels, but the reported localization thresholds use patch-level validation annotations. Single-neuron localization does not preserve all multi-feature slide-classification performance.SAE-1N selects one sparse feature using only slide labels, then uses its patch activation as a tumor-localization signal. A BatchTopK SAE decomposes frozen image embeddings; each feature is max-pooled over a slide and ranked by its tumor-versus-normal association. The selected coordinate is not normalized separately within each slide, unlike softmax attention. The reported thresholded localization still uses annotated validation patches.Official PSB proceedings PDF · Author implementation and supplement

Reading question: Which labels select the feature, tune its threshold, and evaluate localization or slide prediction?

Evidence boundary: Do not hide patch-annotation use in threshold selection or assume improved localization implies better slide AUC. Distinguish retraining after an intervention from a fixed-classifier causal test.

Identify domain features without erasing task-relevant morphology

EXPOSE trains an SAE on pathology embeddings, identifies domain-associated latent coordinates, and masks selected coordinates before downstream prediction. It studies whether this intervention reduces acquisition-domain dependence while retaining useful outcome information.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
EXPOSE: Explainable and Domain-Robust Embeddings from Pathology Vision Foundation Models using Sparse AutoencodersAnja Witte, Maximilian Lennartz, Jan Baumbach, Guido Sauter, Stefan Bonn, Patrick Fuhlert, and Marina Zimmermann2026 · ECCV eXCV Workshop · acceptedIndependently verified on the official eXCV 2026 Proceedings Track list, not ECCV main conference. Scientific review uses arXiv:2608.28191v1; a final typeset proceedings version was not verified. Evaluation is prostate tissue-microarray spots, not a whole-slide benchmark, and includes a thin-section failure case.EXPOSE learns a sparse basis of frozen pathology features, identifies dimensions that distinguish two acquisition domains, and masks selected dimensions before training a relapse classifier. It is an automatic domain-feature selection strategy, not a new image encoder. The evidence concerns prostate tissue-microarray spots and controlled preparation/scanner variants, not unrestricted WSI deployment.Official workshop accepted-paper list · Author implementation · Versioned full manuscript and appendix

Reading question: Does domain-feature suppression preserve prediction in every held-out acquisition condition?

Evidence boundary: The inspected study uses prostate tissue-microarray spots, not a general full-WSI benchmark. Results vary by domain; removing detectable acquisition information does not prove that only nuisance was removed.

Encoder selection: connect cheap proxies to the intended endpoint

Tile-to-slide transferability compares the same frozen encoders on local morphology and slide tasks. Strong aggregate correlations support shortlisting, while learned aggregation and context-heavy tasks can change which model wins. Use this alongside structure-aware feature-selection metrics, not as a replacement for final WSI validation.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
From Patches to Patients: A Study of the Tile-to-Slide Performance Transferability in Digital PathologySofiène Boutaj, Leo Fillioux, Maria Vakalopoulou, Stergios Christodoulidis, and Pierre Marza2026 · MICCAI · peer-reviewedMICCAI 2026 chapter, LNCS 16891, pages 149–159; publisher first-online date is in September 2026, although the volume copyright/bibliographic imprint says 2027. Review uses the inspected publisher PDF, not only arXiv:2606.10778. Across-encoder rank correlation supports shortlisting, not guaranteed selection of the best model for a new clinical cohort.Tile-level linear probing can help shortlist pathology encoders before expensive whole-slide experiments. Across 19 frozen models, aggregate tile performance correlates strongly with aggregate slide performance, but learned attention changes the ranking. The contribution is a benchmark of when a cheaper proxy transfers, not a new encoder, distillation method, or promise that the best tile model wins on every clinical task.Publisher full chapter · Official MICCAI open-access record · Official accepted manuscript · Earlier arXiv manuscript

Reading question: Does the ranking survive the actual slide aggregator and held-out clinical task?

Evidence boundary: Across-model correlation is not guaranteed top-one selection, patient-level validity, or evidence that a distilled student preserves cross-magnification information.

Spatial evidence: separate coherence, molecular agreement, and anatomy

SpaPath-Bench compares frozen image features through a controlled spatial learner and clustering readout on paired histology–ST sections. Its three reference regimes expose different strengths: a smooth partition, a transcriptomic proxy, and an expert anatomical boundary are not interchangeable targets.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Benchmarking Pathology Foundation Models for Spatial Domain UnderstandingBokai Zhao, Yiyang Zhang, Yuanchi Zhu, Hanqing Chao, Long Bai, Tai Ma, Minfeng Xu, Ming Song, and Tianzi Jiang2026 · MICCAI · peer-reviewedMICCAI 2026 chapter, LNCS 16895, pages 35–45, first online 25 September 2026; the volume imprint says 2027. Review uses the inspected publisher PDF. The 42 slides comprise 12 expert-annotated DLPFC slides and 30 slides assessed using transcriptomic clustering proxies. The linked public repository is a prepared-results dashboard, not a verified end-to-end benchmark implementation.SpaPath-Bench asks whether frozen pathology features support meaningful tissue partitions after a controlled spatial readout. It compares 19 encoders on 42 paired H&E–spatial-transcriptomics slides, separating spatial coherence, molecular-partition agreement and expert-label agreement. The evaluated object is an encoder-plus-readout system, not innate global spatial reasoning by a local tile encoder.Publisher full chapter · Official MICCAI open-access record · Official accepted manuscript · Author results dashboard · Dashboard repository and documented limits

Reading question: Which spatial relationship is preserved, and how much is supplied by the readout rather than the encoder?

Evidence boundary: Only 12 DLPFC slides carry expert labels; the other 30 use molecular clustering proxies. Repeated configurations are not independent specimens, and single-slide visual examples are not cohort means.

Complete maintained paper route

Section titled “Complete maintained paper route”

94 canonical readings · 2017–2026· route reviewed through October 2, 2026.

This is the complete structured list for this chapter, ordered chronologically from earlier foundations toward current work. The explanatory ladder above highlights the shortest conceptual route; this list also includes important branches, benchmarks, and current endpoints. It is generated from the same deduplicated catalog used by the AI background registry and My Library, so a future catalog update appears here automatically.

Show all 94 readings in maintained order

  16. #162023IEEE Journal of Biomedical and Health InformaticsPeer reviewedL3

    Deep Multi-Magnification Similarity Learning for Histopathological Image Classification

    Songhui Diao, Weiren Luo, Jiaxin Hou, Ricardo Lambo, Hamas A. AL-kuhali, Hanqing Zhao, Yinli Tian, Yaoqin Xie, Nazar Zaki, and Wenjian Qin

    Peer-reviewed journal article 27(3):1535–1545. DMSL learns paired same-field-of-view patch features at two magnifications and retains both magnification branches at inference; its reported comparison metrics are patch-level, not a WSI endpoint.

    Official artifacts: PubMed record

    Metadata checked

  21. #212024arXivPreprintL3

    Learning biologically relevant features in a pathology foundation model using sparse autoencoders

    Nhat Minh Le, Ciyue Shen, Neel Patel, Chintan Shah, Darpan Sanghavi, Blake Martin, Alfred Eng, Daniel Shenker, Harshith Padigela, Raymond Biju, Syed Ashar Javed, Jennifer Hipp, John Abel, Harsha Pokkalla, Sean Grullon, and Dinkar Juyal

    Review uses arXiv:2407.10785v3, including its supplement. A later NeurIPS 2025 Imageomics workshop paper has the title Evaluating the Utility of Sparse Autoencoders for Interpreting a Pathology Foundation Model; its inaccessible full PDF was not used to replace this version's claims or publication status.

    Official artifacts: Versioned full manuscript · Authors' later Imageomics workshop publication

    Metadata checked

  41. #412025MICCAIPeer reviewedL3

    Unsupervised Anomaly Detection on Preclinical Liver H&E Whole Slide Images Using Graph-Based Feature Distillation

    Lin Li, Lillie Shelton, Thomas Forest, Kyathanahalli Janardhan, Tiffany Jenkins, Michael J. Napolitano, Roujia Wang, David Leigh, Tosha Shah, Grady Earl Carlson, Rajath Soans, and Antong Chen

    Peer-reviewed MICCAI 2025 proceedings paper; evaluated on preclinical rat-liver H&E WSIs. The author project page requests email contact for code access and showed code as coming soon when checked on 2026-08-23.

    Official artifacts: Author project and code-access page

    Metadata checked

  42. #422025Scientific ReportsPeer reviewedL3

    Human Visual Attention-Inspired Knowledge Distillation Underlying Interpretable Computational Pathology

    Muzhou Yu, Zihan Zhong, Xingang Zhou, Yuekun Wang, Tingyu Liang, Jiamin Chen, Hongmin Huang, Junhan Zhou, Dachun Zhao, Bo Lei, Yu Wang, Wenbin Ma, and Kaisheng Ma

    Peer-reviewed open-access journal article 15:42124. Low- and high-magnification language motivates the design, but HVisKD transfers sample- and region-relation features between teacher and student given the same pathology patches; it is not cross-magnification distillation.

    Official artifacts: Open-access article · Official code

    Metadata checked

  43. #432025MICCAIPeer reviewedL3

    Prototype-Based Multiple Instance Learning for Gigapixel Whole Slide Image Classification

    Susu Sun, Dominique van Midden, Geert Litjens, and Christian F. Baumgartner

    Official MICCAI 2025 record and 11-page accepted manuscript checked. PANDA is evaluated as binary tumor classification in this study, not the original six-class ISUP grading task. Artifact concepts are masked before MIL retraining; this is not a universal AUC improvement.

    Official artifacts: Official MICCAI open-access record · Official accepted manuscript · Author implementation · arXiv manuscript

    Metadata checked

  58. #582026NeurIPSAccepted · proceedings pendingL3

    PathView-Bench: Can Multimodal Large Language Models Achieve Fine-grained Multiscale Understanding of Pathology Images?

    Zongyi Chen, Yu Liang, Jie Lin, and Liansheng Wang

    Acceptance independently verified 2026-10-08: official NeurIPS poster 139172 exactly matches the title and four authors and links OpenReview gZJHNK7GHw. The scientific review remains tied to arXiv:2607.28318v1 and its 2026-08-08 full-text audit; this metadata refresh did not re-audit code or dataset availability.

    Official artifacts: Official NeurIPS 2026 acceptance record

    Metadata checked

  60. #602026MICCAIPeer reviewedL3

    Harnessing Adversarial Distillation to Customise Debiased, Disease-Specific Pathology Foundation Models for Breast Cancer

    Zhiwei Chen, Yang Hu, Yuxiang Xiao, Yakun Ju, Tianyang Zhang, Yingxue Xu, Wei Li, Hao Chen, Jens Rittscher, and Kaixiang Yang

    Official MICCAI 2026 proceedings record and accepted-version open-access PDF verified 2026-10-02; Springer DOI and pagination remain pending.

    Official artifacts: Official code · Official MICCAI 2026 proceedings · Official accepted-version PDF

    Metadata checked

  70. #702026arXivPreprintL2–L3

    Paired Uterine Whole-Slide Images and Pathology Reports for Multimodal Computational Pathology

    Han Li, Jingsong Liu, Ayako Ura, Junlin Hou, Zhengyang Xu, Azar Kazemi, Oskar Thaeter, Christian Grashei, Fabian Gülhan, Reza Nasirigerdeh, Xun Ma, Rui Yan, Hao Chen, S. Kevin Zhou, Nassir Navab, Carolin Mogler, and Peter Schüffler

    arXiv preprint submitted 4 July 2026 and revised as v2 on 17 July 2026; no peer-reviewed version was verified on 20 August 2026.

    Official artifacts: Official gated dataset card

    Metadata checked

  71. #712026Medical Image AnalysisPeer reviewedL3

    Streamline Pathology Foundation Model by Cross-Magnification Distillation

    Ziyu Su, Abdul Rehman Akbar, Usama Sajjad, Anil V. Parwani, and Muhammad Khalid Khan Niazi

    Journal record verified through publisher metadata, Crossref and PubMed PMID 42874158 on 26 October 2026: electronic publication 10 October 2026, assigned to volume 115 (January 2027), article 104352. The stable ID retains the 2025 preprint year. Journal full text was unavailable; detailed scientific claims and visuals remain explicitly tied to the previously inspected arXiv v3, not asserted to match the journal version. The journal abstract describes seven tasks and encoder-level throughput, not an end-to-end speedup.

    Official artifacts: Inspected author manuscript (v3) · PubMed journal metadata and abstract · Official code · Official model

    Metadata checked

  72. #722026ECCVAccepted · proceedings pendingL3

    Pathryoshka: Compressing Pathology Foundation Models via Multi-Teacher Knowledge Distillation With Nested Embeddings

    Christian Grashei, Christian Brechenmacher, Rao Muhammad Umer, Jingsong Liu, Carsten Marr, Peter J. Schüffler, and Ewa Szczurek

    ECCV 2026 poster status reverified against the official complete program (poster 4454). No oral or spotlight designation is recorded. A Springer version of record was not verified on 2026-09-12; the existing review retains its declared manuscript revision.

    Official artifacts: Official ECCV record · Official model · Conference manuscript

    Metadata checked

  73. #732026Medical Image AnalysisAccepted · proceedings pendingL3

    RACR-MIL: Rank-aware contextual reasoning for weakly supervised grading of squamous cell carcinoma using whole slide images

    Anirudh Choudhary, Mosbah Aouad, Krishnakant Saboo, Angelina Hwang, Jacob Kechter, Blake Bordeaux, Puneet Bhullar, David DiCaudo, Steven Nelson, Nneka Comfere, Emma Johnson, Olayemi Sokumbi, Jason Sluzevich, Leah Swanson, Dennis Murphree, Aaron Mangold, and Ravishankar Iyer

    The complete arXiv v3 manuscript identifies Medical Image Analysis as the journal and the abstract record reports acceptance; a final DOI or journal issue record was not yet available when checked on 30 August 2026.

    Metadata checked

  76. #762026NeurIPSAccepted · proceedings pendingL3

    VIPER: An Expert-Curated Benchmark for Vision-Language Models in Veterinary Pathology

    Luca L. Weishaupt, Simone de Brot, Javier Asin, Llorenç Grau-Roma, Nic G. Reitsam, Andrew H. Song, Dongmin Bang, Stefan T. Kaluziak, Long Phi Le, Jakob Nikolas Kather, Faisal Mahmood, and Guillaume Jaume

    Acceptance independently verified 2026-10-08 through official NeurIPS poster 139742: exact title and matching 12-author list, linked OpenReview 7986DZ9rPe. Scientific review and artifact checks remain tied to arXiv:2608.26382v1 and the 2026-08-30 audit, not a refreshed full-text/code/data inspection. VIPER evaluates rat toxicologic regions rather than whole-slide images; it is an ROI benchmark, not a WSI benchmark.

    Official artifacts: Official evaluation code and documentation · Official benchmark dataset · Official NeurIPS 2026 acceptance record

    Metadata checked

  77. #772026Findings of EMNLPAccepted · proceedings pendingL3

    MVC-Bench: Benchmarking Calibration of Medical Vision-Language Models

    Ashshak Sharifdeen, Shihab Aaqil Ahamed, Ufaq Khan, Muhammad Akhtar Munir Sujair Ibrahim, Mohamed Rafeek Mareer Ahamed, Yutong Xie, Imran Razzak, and Muhammad Haris Khan

    The arXiv record reports acceptance to Findings of EMNLP 2026. No final ACL Anthology record was available when checked on 30 August 2026, so claims were verified against arXiv v1 and the official repository.

    Official artifacts: Official implementation

    Metadata checked

  78. #782026arXivPreprintL3

    Benchmarking Vision-Language Models for Automated Pathology Diagnosis and Report Generation

    Yumi Lee, Harim Oh, Hyoryung Kim, Minji Kim, Eunsu Kim, Hyeseong Lee, Junya Fukuoka, Andrey Bychkov, Jijgee Munkhdelger, Rajiv Kumar Kaushal, Ayushi Sahay, Rajni Yadav, Bharathi Prabakaran, Sulen Sarioglu, Serdar Balcı, Ilknur Turkmen, Yuri Tolkach, Christian Harder, Julian Westerdorf, Reinhard Buettner, Audun Ljone Henriksen, Sepp De Raedt, Byung Hyun Lee, Sungjin Lim, Joohoon Lee, Gwanghyun Kim, Se Young Chun, Suryakant Singh, Saarthak Kapse, Prateek Prasanna, Kyung A Kim, Yousun Kang, Sehwan Yoo, Sungman Hong, Shubham Innani, Michael Feldman, Spyridon Bakas, Ujjwal Baid, Prasad Dutande, Suhas Gajare, Bhakti Baheti, Serkan Sökmen, Ece Tuğba Cebeci, Ahmet Halıcı, Musa Balcı, Kardelen Peçenek, Srividhya Sainath, Kyongseok Jang, Messi H. J. Lee, Noorul Wahab, Bodong Du, Jiaming Zhang, Qixiang Zhang, Jang-Hwan Choi, and Sangjeong Ahn

    arXiv v1 submitted 1 September 2026 reports the results of the MICCAI REG 2025 challenge. No final peer-reviewed proceedings or journal record was verified on 2 September 2026. The approximately 10,500 institutional WSI-report pairs are not a generally downloadable public dataset; the released repositories expose benchmark information and team artifacts.

    Official artifacts: Official challenge repository · Official winning-system repository · Official challenge site

    Metadata checked

  81. #812026Engineering Applications of Artificial IntelligencePeer reviewedL3

    Cross-Stain Knowledge Distillation for Low-Cost Lung Cancer Programmed Death Ligand-1 Assessment With Multi-Granularity Multiple Instance Learning

    Yi Shi, Chong Ge, Fang Zhao, Anli Zhang, Ao Li, Haibo Wu, and Minghui Wang

    Peer-reviewed journal article, volume 167, article 113762. IHC is privileged teacher information during training; deployment uses the H&E student, but its macro- and micro-granularity branches remain. No official code or model release was verified on 2026-09-02.

    Metadata checked

  82. #822026Scientific ReportsPeer reviewedL3

    KDTViT Knowledge Distillation, Transfer Learning and Transformer Based Deep Learning Framework for Efficient Histopathology Image Classification

    Aarv Mankodi, Praveen Kumar Shukla, and Hitesh Tekchandani

    Peer-reviewed open-access journal article 16:20721. A MobileViT-XS is sequentially adapted from 40× through 400× using the previous checkpoint as teacher; BreakHis consists of biopsy images rather than WSIs, and the same-capacity teacher/student design targets retention rather than model compression.

    Official artifacts: Open-access article

    Metadata checked

  83. #832026Computers & Electrical EngineeringPeer reviewedL3

    CMA-MIL: A Cross-Magnification Attention-Based Multiple Instance Learning Framework for Renal Cell Carcinoma Subtyping Using Whole-Slide Images

    Rashika Bagri and Ankit Rajpal

    Peer-reviewed journal article, volume 135, article 111204. The weakly supervised model uses 5×, 10×, and 20× streams and retains their cross-attention fusion at inference; it is not a low-magnification-only student. No official artifact was verified on 2026-09-02.

    Metadata checked

  84. #842026Medical Image AnalysisPeer reviewedL3

    M2OTCA: Multiple-Magnification Optimal Transport-Based Cross-Attention Learning for Whole Slide Image Classification

    Zhonghang Zhu, Erik Meijering, and Liansheng Wang

    Peer-reviewed journal article, volume 112, article 104082. Region structural prototypes reduce the cost of optimal transport across 5×, 10×, and 20×; all three magnifications feed magnification-specific predictions and remain part of inference. No official artifact was verified on 2026-09-02.

    Official artifacts: PubMed record

    Metadata checked

  85. #852026arXivPreprintL3

    One Model to Magnify Them All: Efficient Scale-Invariant Histopathology via Conditional Normalization and Continuous Magnification Training

    Agnieszka Florkowska, Henning Müller, and Marek Wodzinski

    arXiv v1 submitted 10 August 2026; no peer-reviewed venue was verified on 2026-09-02. A single CNN is conditioned on physical pixel size and trained over a continuous magnification range; each prediction still consumes one patch scale and the method is neither MIL nor knowledge distillation.

    Official artifacts: Official code

    Metadata checked

  86. #862026ECCVAccepted · proceedings pendingL3

    SurvMILKD: A Weakly Supervised Survival Analysis Framework for Multi-Teacher Knowledge Distillation using Pathology Foundation Models

    Mayur Mallya, Ali Khajegili Mirabadi, Hossein Farahani, and Ali Bashashati

    The official ECCV 2026 poster and conference-hosted manuscript verify acceptance; a Springer version of record was not verified. The official repository was a code-coming-soon placeholder on 2026-09-12. All foundation encoders remain frozen: task-specific MIL adapters and translation modules learn, not the UNI backbone.

    Official artifacts: Official ECCV 2026 Poster · Official repository — code pending

    Metadata checked

  87. #872026ECCVAccepted · proceedings pendingL3

    TaxoMIL: Taxonomy-Constrained Learning for Hierarchical Whole Slide Image Analysis

    Chaeyeon Lee, Khang Nguyen Quoc, Jinsol Song, Yosep Chong, Kwangil Yim, and Jin Tae Kwak

    Official ECCV 2026 poster and conference-hosted manuscript verify acceptance; no Springer version of record was verified. Author spelling follows the paper rather than the abbreviated program entry. Output is coarse/fine diagnostic label text, not full clinical reports. GastWSI is private.

    Official artifacts: Official ECCV 2026 Poster · Official implementation · arXiv manuscript

    Metadata checked

  89. #892026arXivPreprintL3

    What Makes High-Magnification Knowledge Transferable? A Study of Cross-Resolution Distillation in Whole-Slide Imaging

    Zhiyuan Yang, Jiahao Cheng, and Mahdi S. Hosseini

    Public arXiv v2 (2026-10-01) verified 2026-10-02. The authors state ICLR 2027 under review; no acceptance inferred. Only public source material was used. A diagnostic study of target utility, predictability and downstream benefit, not a new universally superior distillation architecture.

    Official artifacts: Reviewed manuscript revision

    Metadata checked

  90. #902026Pacific Symposium on BiocomputingPeer reviewedL3

    Patch-level phenotype identification via weakly supervised neuron selection in sparse autoencoders for CLIP-derived pathology embeddings

    Keita Tamura, Yao-zhong Zhang, Yohei Okubo, and Seiya Imoto

    PSB 2026 volume 31, pages 708–721, with source supplement checked. Neuron discovery uses slide labels, but the reported localization thresholds use patch-level validation annotations. Single-neuron localization does not preserve all multi-feature slide-classification performance.

    Official artifacts: Official PSB proceedings PDF · Author implementation and supplement

    Metadata checked

  91. #912026ECCV eXCV WorkshopAccepted · proceedings pendingL3

    EXPOSE: Explainable and Domain-Robust Embeddings from Pathology Vision Foundation Models using Sparse Autoencoders

    Anja Witte, Maximilian Lennartz, Jan Baumbach, Guido Sauter, Stefan Bonn, Patrick Fuhlert, and Marina Zimmermann

    Independently verified on the official eXCV 2026 Proceedings Track list, not ECCV main conference. Scientific review uses arXiv:2608.28191v1; a final typeset proceedings version was not verified. Evaluation is prostate tissue-microarray spots, not a whole-slide benchmark, and includes a thin-section failure case.

    Official artifacts: Official workshop accepted-paper list · Author implementation · Versioned full manuscript and appendix

    Metadata checked

  92. #922026MICCAI MI4MedFM WorkshopAccepted · proceedings pendingL3

    PATH-DISSECT: Interpreting Concept Representation and Classifier Reliance in Pathology Foundation Models with Sparse Autoencoders

    Mayur Parvatikar, Xingjian Li, and Min Xu

    Official MICCAI 2026 Satellite Events page supplies the 11-page camera-ready; Springer DOI and final pages were pending at inspection. This is MI4MedFM workshop work on tissue patches, not main-conference or patient-level WSI outcome validation. Causal claims concern interventions on a frozen encoder plus linear readout, not biological causation.

    Official artifacts: Official camera-ready PDF · Author implementation · Workshop OpenReview record

    Metadata checked

  93. #932026MICCAIPeer reviewedL3

    From Patches to Patients: A Study of the Tile-to-Slide Performance Transferability in Digital Pathology

    Sofiène Boutaj, Leo Fillioux, Maria Vakalopoulou, Stergios Christodoulidis, and Pierre Marza

    MICCAI 2026 chapter, LNCS 16891, pages 149–159; publisher first-online date is in September 2026, although the volume copyright/bibliographic imprint says 2027. Review uses the inspected publisher PDF, not only arXiv:2606.10778. Across-encoder rank correlation supports shortlisting, not guaranteed selection of the best model for a new clinical cohort.

    Official artifacts: Publisher full chapter · Official MICCAI open-access record · Official accepted manuscript · Earlier arXiv manuscript

    Metadata checked

  94. #942026MICCAIPeer reviewedL3

    Benchmarking Pathology Foundation Models for Spatial Domain Understanding

    Bokai Zhao, Yiyang Zhang, Yuanchi Zhu, Hanqing Chao, Long Bai, Tai Ma, Minfeng Xu, Ming Song, and Tianzi Jiang

    MICCAI 2026 chapter, LNCS 16895, pages 35–45, first online 25 September 2026; the volume imprint says 2027. Review uses the inspected publisher PDF. The 42 slides comprise 12 expert-annotated DLPFC slides and 30 slides assessed using transcriptomic clustering proxies. The linked public repository is a prepared-results dashboard, not a verified end-to-end benchmark implementation.

    Official artifacts: Publisher full chapter · Official MICCAI open-access record · Official accepted manuscript · Author results dashboard · Dashboard repository and documented limits

    Metadata checked