Skip to content
Computational Pathology Reading Roadmap

WSI engineering and quality control

A whole-slide image (WSI) is not a large JPEG. It is a multiresolution acquisition of a physical specimen, usually stored in a vendor format with scanner metadata and a pyramid of image levels. Every downstream result depends on how that specimen became pixels, how pixels became tiles, and which tiles were allowed into the model.

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

This chapter covers the engineering layer that is often hidden behind “we extracted patches.” It connects data lineage, coordinates, quality control, stain variation, sampling, storage, and reproducible feature extraction. The WSI Data Atlas covers dataset access; this page covers what to do after a slide is obtained.

Learning outcomes

Section titled “Learning outcomes”

After this route, you should be able to:

  1. trace a prediction from patient to case, block, slide, region, tile, and model input;
  2. convert between level-zero pixels, pyramid-level pixels, microns, and tile indices;
  3. design QC gates without silently changing the study population;
  4. compare stain normalization with stain augmentation as experimental interventions;
  5. build an immutable tile or embedding manifest that another researcher can audit; and
  6. distinguish computational throughput from scientific reproducibility.

1. Preserve the specimen hierarchy

Section titled “1. Preserve the specimen hierarchy”

The safe unit of analysis is rarely the image file.

patient
└── case / encounter
    └── specimen
        └── block
            └── glass slide
                └── WSI acquisition
                    └── tissue region
                        └── tile

One patient can have multiple cases, blocks, slides, stains, recuts, and scanner acquisitions. Splitting at tile or slide level can therefore leak near-duplicate biology across train and test sets. Assign the patient—or the clinically justified higher unit—to a split before tissue detection or tiling.

Keep a private linkage table for protected identifiers and a research manifest with stable pseudonymous IDs. Do not infer independence merely because filenames differ.

2. Understand the image pyramid

Section titled “2. Understand the image pyramid”

For each slide, record:

  • level-zero width and height;
  • downsample factor for every pyramid level;
  • microns per pixel (MPP) on both axes when available;
  • scanner vendor/model, objective metadata, and acquisition date;
  • compression and color profile;
  • stain and specimen type; and
  • whether the file is original, converted, rescanned, or derived.

If d_l is the downsample factor of level l, a coordinate at that level maps to level zero approximately as:

(x_0, y_0) = (d_l × x_l, d_l × y_l)

Do not assume “20×” means the same physical resolution across scanners. Prefer MPP, verify metadata against known slide properties when possible, and store the actual read level plus any software resampling. Anisotropic MPP and non-integer downsample factors are possible.

Coordinate contract

Section titled “Coordinate contract”

Choose one canonical coordinate system—normally level-zero pixels—and require every annotation, tissue mask, tile, heatmap, and cell object to carry:

slide_id: stable-research-id
level0_xywh: [x, y, width, height]
read_level: 1
read_level_downsample: 4.0001
source_mpp_xy: [0.247, 0.247]
model_mpp_xy: [0.5, 0.5]
resampling: area
tile_size_px: [224, 224]

This prevents a common failure: an apparently correct overlay displaced because an annotation and a heatmap use different pyramid levels.

3. Make quality control explicit

Section titled “3. Make quality control explicit”

QC is a sequence of decisions, not one binary classifier.

Layer Examples Recommended action What to log
File and metadata unreadable file, missing levels, impossible MPP quarantine; repair or reacquire reader version, exception, checksum
Acquisition blur, striping, stitching, missing tissue, scanner debris mask local defects or exclude by a prespecified rule defect type, area, reviewer, threshold
Preparation folds, chatter, bubbles, pen, precipitate, crushed tissue preserve biologically valid tissue; mask technical artifact artifact mask and reason
Cohort stain/scanner/site outlier, unusual dimensions inspect for batch shift; do not discard automatically cohort metrics and decision
Model input background, whitespace, duplicate/empty tiles deterministic filter mask version, tissue fraction, tile count

An outlier can be an error, a rare valid phenotype, or a new deployment domain. Automatic removal without review can make a benchmark easier while narrowing the claimed population. Report the number of patients, slides, and tissue area removed at every gate.

Useful starting tools include HistoQC, GrandQC, and viewer-based review in QuPath. A tool score is not a universal acceptance threshold; calibrate the rule to the specimen, stain, task, and intended use.

4. Tissue detection and tile sampling

Section titled “4. Tissue detection and tile sampling”

Tissue detection establishes the population of instances seen by a slide model. Keep the tissue mask as a versioned artifact and visually inspect difficult cases: fat, mucin, pale stroma, necrosis, fragmented biopsies, dark IHC, and pen markings.

Sampling policies answer different questions

Section titled “Sampling policies answer different questions”
Policy Advantage Main risk
Dense grid complete, easy to reproduce large compute and many redundant/background tiles
Tissue-threshold grid efficient default threshold can remove pale or tiny tissue
Random tissue sample scalable and regularizing seed- and budget-dependent coverage
ROI/tumour-first focuses on a known compartment imports an annotation model and its errors
Hard or attention-guided allocates compute to uncertain evidence feedback loop can amplify early shortcuts
Multi-scale paired joins architecture and cytology registration, context, and memory complexity

State tile size in both pixels and physical units, overlap/stride, edge policy, background threshold, magnification/MPP, maximum tiles, selection seed, and whether sampling changes between training and evaluation. For a slide label, never select tiles using information unavailable at deployment.

5. Treat stain handling as a hypothesis

Section titled “5. Treat stain handling as a hypothesis”

Stain variation comes from fixation, processing, section thickness, reagents, protocol, scanner, and color management. Three broad strategies are:

  1. leave color unchanged and measure generalization;
  2. augment stain/color during training to learn invariance; or
  3. normalize to a target appearance before training and inference.

Normalization is not automatically safer. It can add artifacts, make performance depend on the reference slide, or remove clinically useful variation. Compare raw, augmentation-only, and normalization-plus-augmentation pipelines on held-out sites. Fit or choose any reference using training data only.

Classic starting points are Macenko et al. (2009) and the structure-preserving method of Vahadane et al. (2016). Then read the multi-centre comparison by Tellez et al. (2019) before adopting a normalizer by convention.

6. Build an immutable feature pipeline

Section titled “6. Build an immutable feature pipeline”

For foundation encoders and MIL, feature extraction is often the expensive boundary. Key every output by all inputs that can change it:

feature_set_id =
  hash(slide checksum,
       tissue-mask version,
       coordinate manifest,
       read MPP and resampling,
       tile transform,
       encoder release and weights checksum,
       inference code version)

Store coordinates beside embeddings. Record tensor dtype, shape, pooling token, normalization, batch size, device/software versions, and failures. Never overwrite an old feature set under the same name after changing the encoder or preprocessing.

At minimum, produce these artifacts:

  • slides: one row per acquisition and its lineage;
  • qc: slide- and region-level decisions;
  • tiles: canonical coordinates and sampling metadata;
  • features: pointer, shape, dtype, encoder identity, checksum;
  • splits: immutable patient/case assignments; and
  • failures: missing/corrupt slides and processing exceptions.

7. Systems and standards reading route

Section titled “7. Systems and standards reading route”

Read in order within each stage. Years are years of the version of record.

Stage A — WSI representation and annotation

Section titled “Stage A — WSI representation and annotation”
Order Primary source Year / status Read for
1 OpenSlide: A Vendor-Neutral Software Foundation for Digital Pathology 2013 · Journal of Pathology Informatics vendor formats, random-access regions, pyramids, metadata, and why interoperability is difficult
2 QuPath: Open Source Software for Digital Pathology Image Analysis 2017 · Scientific Reports annotation, measurements, scripting, and reproducible human-in-the-loop analysis
3 Implementing the DICOM Standard for Digital Pathology 2018 · Journal of Pathology Informatics tiled total-pixel-matrix objects, specimen metadata, compression trade-offs, and a multivendor DICOM/DICOMweb implementation; keep the living standard version-pinned
4 OME-NGFF: a next-generation file format for expanding bioimaging data-access strategies 2021 · Nature Methods Zarr-backed chunked multiscale arrays, shared metadata, labels, cloud access, and reference implementations; keep the living specification version-pinned
5 TIAToolbox as an End-to-End Library for Advanced Tissue Image Analytics 2022 · Communications Medicine modular readers, patch prediction, WSI inference, and reusable pipeline interfaces
6 SpatialData Nature Methods, published online 2024 / volume 2025 explicit transformations and shared coordinate systems across images, labels, points, shapes, tables, and spatial assays
7 LazySlide Nature Methods 2026, peer reviewed interoperable WSI access, foundation-model features, cell segmentation, spatial domains, and SpatialData integration; it is a research framework, not a clinically validated system

Stage B — QC and cohort shift

Section titled “Stage B — QC and cohort shift”
Order Primary source Year / status Read for
8 HistoQC 2019 · JCO Clinical Cancer Informatics interpretable QC modules, artifact masks, and cohort outlier views
9 The Impact of Site-Specific Digital Histology Signatures on Deep Learning Model Accuracy and Bias 2021 · Nature Communications why site can remain predictable after common color controls
10 Efficient Quality Control of Whole Slide Pathology Images with Human-in-the-Loop Training 2023 · Scientific Reports foreground/QC learning and the role of targeted human correction
11 GrandQC 2024 · npj Precision Oncology multi-class artifact segmentation and multi-site/scanner QC benchmarking
12 Enterprise automated WSI QC deployment Journal of Pathology Informatics 2026, peer reviewed operational evidence over 94,995 WSIs and 11,085 cases; retain single-center, commercial-system, and non-randomized limitations

Stage C — stain and pipeline interventions

Section titled “Stage C — stain and pipeline interventions”
Order Primary source Year / status Read for
13 A Method for Normalizing Histology Slides for Quantitative Analysis 2009 · ISBI optical-density space and stain-vector estimation
14 Structure-Preserving Color Normalization and Sparse Stain Separation 2016 · IEEE TMI sparse stain decomposition and morphology preservation assumptions
15 Quantifying the Effects of Data Augmentation and Stain Color Normalization 2019 · Medical Image Analysis controlled multi-organ/multi-lab comparison and practical experimental design
16 StainNet: A Fast and Robust Stain Normalization Network 2021 · Frontiers in Medicine learned pixel mapping, speed, and normalization artifacts
17 Physical scanner color calibration Modern Pathology 2025, peer reviewed separate acquisition-device calibration from computational stain normalization—they address different shift sources

Stage D — diagnostic workflow validation

Section titled “Stage D — diagnostic workflow validation”
Order Primary source Year / status Read for
18 CAP WSI validation guideline online 2021 / issue 2022, active guideline validation of the scanner–viewer–display diagnostic workflow; this is distinct from validating an AI algorithm

Stage E — resolution and systems-cost interventions

Section titled “Stage E — resolution and systems-cost interventions”
Order Primary source Year / status Read for
19 Resolution-Based Distillation for Efficient Histology Image Classification 2021 · Artificial Intelligence in Medicine, peer reviewed the patch-level precursor: a high-resolution classifier teaches an architecturally matched low-resolution student through outputs and spatial feature maps, with an unlabeled transfer set
20 Differentiable Zooming for MIL (ZoomMIL) 2022 · ECCV, peer reviewed coarse-to-fine learned evidence acquisition; high-resolution regions are still read after selection, so this is not low-resolution-only deployment
21 Attention to Detail: Inter-Resolution Knowledge Distillation 2023 · EUSIPCO, peer reviewed Grad-CAM-based attention transfer for lower-resolution prostate histology grading; a patch classifier, not a WSI MIL system
22 A Graph-Based Multi-Scale Approach With Knowledge Distillation for WSI Classification 2024 · IEEE TMI, peer reviewed DINO features at 5×/10×/20×, spatial graphs, per-scale MIL, and cross-scale self-distillation; all scale branches remain part of inference
23 Reducing Histopathology Slide Magnification Improves the Accuracy and Speed of Ovarian Cancer Subtyping 2024 · ISBI, peer reviewed a controlled 1.25×–40× magnification ablation showing that lower scale can itself be a strong baseline; there is no teacher–student distillation
24 CEMIL 2025 · WACV, peer reviewed frozen instructor modules score all pre-extracted 20× patch embeddings before a subset learner is distilled; report full extraction and instructor scoring beside the paper’s learner-side FLOP and inference-time reductions
25 PatHME 2025 · IEEE Access, peer reviewed prompt-guided 5×/20× transfer and multiple foundation-model experts followed by hierarchical multiscale attention; both scale streams are retained
26 H0-mini 2025 · MICCAI, peer reviewed a different efficiency axis: distil a large pathology foundation encoder into a compact checkpoint and test scanner/stain robustness, without changing WSI magnification
27 GraphTox 2025 · MICCAI, peer reviewed teacher–student feature discrepancy on multiresolution liver-WSI graphs for unsupervised anomaly detection; both same-architecture branches and the graph pipeline remain at inference, so this is not compression
28 XMAG 2025 · preprint mean of sixteen 20× UNI2 tile-CLS targets supervises the 5× student CLS, while the individual teacher CLS targets align with sixteen pooled regions of the student’s patch-token grid; only the 5× student is required downstream; use the official code/model and raw timing table rather than inconsistent prose speed ratios
29 Magnification-Aware Distillation (MAD) 2025 · preprint label-free alignment of registered low-magnification context and high-magnification detail for scale-consistent embeddings; this is representation invariance, not yet a universal WSI speed claim
30 MAG-GLTrans 2026 · Cancer Research, peer reviewed registered same-field magnification alignment plus a low-magnification global–local WSI model; the paper reports up to 10.7× compute and more than 20× communication/storage savings within its experiments
31 LRMIL 2026 · MICCAI, peer reviewed; official proceedings verified align one 5× field with sixteen 20× patches, then distil MIL logits, soft attention, and hard instances; deployment retains only the 5× encoder and MIL head; inspect the author code
32 LanGuSTE 2026 · substantially revised preprint language-derived task concepts and cross-scale visual prompts score a 5× view before selected regions are encoded at 20×; include low-resolution scoring, retained high-resolution area, miss rate, and 20× extraction/encoding in the cost ledger
33 Decoupled I/O-Dominant Pipelines for Large-Scale WSI Embedding Extraction 2026 · preprint separate patch staging, embarrassingly parallel embedding, and sharded vector-database ingestion so storage and compute can scale independently; no reusable official implementation was verified at the cutoff

Do not compare three deployment contracts as one method family

Section titled “Do not compare three deployment contracts as one method family”
Contract High-resolution path at inference Representative papers Correct question
Low-only after transfer removed Resolution KD, XMAG, MAG-GLTrans, LRMIL how much registered high-detail information survives in the low-resolution student?
Retained multiscale always retained graph multi-scale KD, PatHME does complementary scale fusion justify the additional extraction, storage, and encoding cost?
Selective high-resolution acquisition retained for chosen regions ZoomMIL, HDMIL, LanGuSTE does a coarse view—possibly guided by language concepts—find rare decisive evidence before the high-resolution acquisition budget is exhausted?
Selection after full feature extraction all 20× embeddings are produced before selection CEMIL does a smaller learner bag justify instructor scoring when extraction cost is still paid for every patch?
Decoupled embedding infrastructure depends on the upstream evidence policy I/O-dominant WSI pipeline which stage limits end-to-end throughput: slide storage, decoding, transfer, encoder, serialization, or vector-database ingestion?

LRMIL’s timing is a within-study result because its paper does not report hardware. Likewise, XMAG’s prose contains speed ratios that do not all agree with its raw timing table. Treat reported seconds as source-specific measurements, not portable constants. Every comparison must record physical MPP, registered field of view, interpolation, patch count, slide IO, extraction, encoding, aggregation, cache storage, communication, hardware, and accuracy. Scanner labels such as “5×” and “20×” alone do not establish equivalent fields.

LanGuSTE is not conventional knowledge distillation and not 5×-only deployment: selected 20× regions remain in its inference path. The decoupled I/O paper changes the systems graph rather than the diagnostic model. These distinctions keep a model speed claim from being confused with complete WSI turnaround time.

8. Minimum reproducibility checklist

Section titled “8. Minimum reproducibility checklist”

Before comparing models, verify:

  • patient/case split precedes tiling and feature extraction;
  • original slide checksum and acquisition metadata are retained;
  • exact reader, QC, tissue-mask, and annotation versions are recorded;
  • every tile has a canonical coordinate, physical scale, and sampling reason;
  • exclusions are reported at patient, slide, and region levels;
  • stain intervention is fit on training data and ablated;
  • encoder weights and preprocessing are release-pinned;
  • embeddings are linked to coordinates and immutable provenance;
  • a random sample of masks, tiles, and overlays is visually audited; and
  • preprocessing failures remain visible rather than silently disappearing.

9. Reproduction project

Section titled “9. Reproduction project”

Choose 30–50 public WSIs from at least two sites or scanners in the WSI Data Atlas. Build one manifest and compare:

  1. raw color versus stain augmentation versus normalization;
  2. two tissue-mask policies;
  3. dense versus budgeted random sampling; and
  4. one frozen encoder with identical downstream evaluation.

Measure retained tissue area, tile-count distribution, QC failure rate, runtime, storage, internal performance, and external-site performance. The deliverable is not only a score: it is the manifest, overlay gallery, exclusion flow, and a written analysis of which engineering decision changed the scientific population.

Continue with cell and tissue modeling for dense labels, MIL and whole-slide learning for weak slide labels, or foundation encoders for reusable representations.

Spatial-token retention for slide pretraining

TokRet uses greedy feature diversity to retain many tile-level representatives; PathTokScale adds sparse slide attention and DINOv2-style pretraining. Compare token selection alone against the complete additional training recipe.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Scaling Whole-Slide Pathology Foundation Model Pretraining with Billions Off-the-Shelf TokensHonglin Li, Zhongyi Shui, Chenglu Zhu, and Lin Yang2026 · ECCV · acceptedECCV 2026 acceptance and the manuscript are verified from the official program. No Springer version of record or author-owned public code/checkpoint was verified. Section 4.1 explicitly combines HistAI with about 10,000 private WSIs; do not describe pretraining as public-only. PathTokScale and PathTokenScaling both occur in the manuscript.TokRet retains diverse spatial tokens from a frozen tile encoder instead of reducing every tile to its CLS token. PathTokScale then pretrains a sparse slide transformer over these richer token sets. The scientific question is how much useful tissue information can be retained per unit of slide-model computation, rather than how to enlarge the tile backbone.Official ECCV 2026 Poster

Reading question: Do retained tokens preserve rare tissue evidence, and how much computation, private data and pretraining are actually required?

Evidence boundary: The corpus includes private WSIs; token-similarity coverage is not a guarantee of diagnostic fidelity or low-magnification equivalence.

Tissue distillation: masks, features, hierarchy and relations

Compare staged pixel/feature KD from Virchow2, PAGET’s hierarchical multi-teacher pseudo-labels, and ConStruct’s transfer of SegFormer structural affinities into adapted CONCH features for prototype-guided segmentation. Their annotation contracts and deployed components differ.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Computationally Efficient Pathology Segmentation using Knowledge Distillation from Foundation ModelsJiaqi Lv, Yijie Zhu, Saki Okada, Maxwell Stanley Renna, Robert Goldin, Stefan Antonowicz, and Shan E Ahmed Raza2026 · arXiv · preprintarXiv v1 (2026-09-03), submitted to Medical Image Analysis according to the authors, not accepted. Verified methods and selected results on 2026-10-02. TIAToolbox is the stated release destination; availability of these exact weights was not independently confirmed. Patch-throughput measurements are not end-to-end WSI latency.Transfer Virchow2 segmentation knowledge into SegFormerB5 or EUNet-M using a staged combination of pixel predictions and global features. The deployed model is a compact segmentation student, not the foundation-model teacher. This changes model cost at fixed patch input, not optical magnification.Versioned manuscript · TIAToolbox — stated release destination

Reading question: Which supervision remains observable at student resolution, and which teacher or post-processing components still run at inference?

Evidence boundary: PAGET-H retains HoverNet. ConStruct’s residual adapters are not residual-error targets. Pixel segmentation, patch throughput and patient-level WSI outcomes are distinct evidence.

Case-level composites and diagnostic text transfer

WILSON combines actual multi-magnification tissue views into composite images and studies patient-level embedding and text-generation variants. Separate image-embedding, report-distillation and generative branches.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
WILSON - a pathology foundation model framework for patient-level analysis and diagnostic text generationSaghir Alfasly, Wataru Uegami, Sobhan Hemati, Wenchao Han, Xiaojia Tang, Kevin Thompson, Daniel Stone, Ghazal Alabtah, Saba Yasir, Michael R. Lucas, Eric W. Klee, Cheryl L. Willman, Judy C. Boughey, Matthew P. Goetz, Krishna R. Kalari, and H. R. Tizhoosh2026 · arXiv · preprintarXiv v1 submitted 2026-09-20. No peer-reviewed record was verified. The manuscript says model release is planned after publication and institutional clearance; Mayo training WSIs/reports are not publicly downloadable.WILSON makes a slide or a multi-slide patient case into a compact multi-magnification composite image, then learns one visual representation aligned with pathology-report supervision. It combines an organ-aware region-selection pipeline, a convolutional encoder, and alternative dense-text, keyword or CLIP alignment branches; a CoCa-style decoder adds caption generation. The central idea is to change the computational unit from thousands of separately encoded tiles to a structured image summary.Full manuscript and supplementary material

Reading question: What images and reports are needed at training and deployment, and which compute path was measured?

Evidence boundary: The system still reads high-power tissue; analytic embedding TFLOPs are not end-to-end speed. It is a preprint with restricted training data and a planned model release.

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.

Condense a slide distribution, not its acquisition magnification

NICER fits adaptive nonparametric prototype sets to each slide’s UNI features using distribution matching. It studies task-agnostic bag condensation, with separate experiments for condensed-training/full-bag testing and condensed training plus testing.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Nonparametric Distribution Matching for Self-Supervised Whole-Slide Image CondensationDuong M. Nguyen, Trong Nghia Hoang, Hang Thi Nguyen, Thanh Trung Huynh, Phi Le Nguyen, and Minh N. Do2026 · NeurIPS · acceptedAcceptance independently verified by NeurIPS 2026 poster 153235 and author repository. Review uses arXiv:2610.00678v1, submitted 2026-09-30. Repository contained only a README on 2026-10-08. Feature condensation at 20× is not low-magnification encoder distillation; Table 1 tests on full feature bags.NICER condenses a whole slide's large frozen-feature bag into a smaller, slide-adaptive set of learned prototypes. Its organizing principle is preserving the distribution seen by the downstream encoder representation, not preserving arbitrary pixel similarity or forcing every slide into the same fixed prototype count. Local pattern exploration preserves diverse features; nonparametric concept condensation then merges redundancy. This is feature/data condensation, not distilling a large encoder into a smaller student.Official NeurIPS 2026 program · Versioned full manuscript · Author repository (README only at audit)

Reading question: Which information survives prototype condensation, and does the measured saving include prototype optimization and feature extraction?

Evidence boundary: The acquisition remains nonoverlapping 256×256 tiles at 20× (0.5 µm/pixel). Table 1 is not a compressed-inference benchmark; feature prototypes are not low-power image tiles or distilled encoder weights.

Surprise-guided navigation and reusable slide memory

PathNavigate uses a low-power map to guide higher-power evidence collection, reusing scan statistics during answer synthesis. Predictive surprise directs exploration while the answering model reads selected tissue evidence; local high-power memory is separate from the global scan memory.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
PathNavigate: A Training-Free Pathology Agent with Surprise-Guided Scan and Shared Slide Memory for Whole-Slide Image VQAChunze Yang, Qidong Liu, Wenjie Zhao, Yue Tang, Jiusong Ge, Di Zhang, Jiashuai Liu, Lei Wu, Junbo Lu, Ni Zhang, Xian Wu, Zeyu Gao, and Chen Li2026 · NeurIPS · acceptedOfficial NeurIPS 2026 poster 154224 verifies acceptance; program title shortens 'Whole-Slide Image VQA' to 'Whole-Slide VQA'. Review uses arXiv:2605.23559v1, not an uninspected camera-ready. Backbone-frozen navigation includes test-time memory optimization and high-magnification evidence; not 5×-only deployment.PathNavigate first establishes what is unusual within the current slide, then asks which of those regions is relevant to the question. A frozen pathology encoder feeds a small per-slide reconstruction memory; its pre-update gradient norm proposes regions, PLIP reranks that pool, and frozen perception and language models answer from selected high-magnification evidence. The same scan statistics provide compact answer-time context. Training-free here means no task-specific backbone training, not absence of test-time optimization.Official NeurIPS 2026 program · Versioned full manuscript · Official inference implementation

Reading question: How much of the answer requires newly sampled high-power tissue, and what does the shared scan summary contribute beyond local evidence?

Evidence boundary: Training-free refers to frozen backbones, not zero optimization: the memory MLP learns online. Released nominal magnifications depend on a hard-coded native 40× assumption; verify scanner metadata. This is not 5×-only inference.

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.

Edge encoders: contrastive multi-teacher compression

MuCoDi caches fixed pathology-teacher vectors and trains compact students to match tile identity contrastively. MuCoEdge is evaluated with frozen-encoder MIL and single-tile CPU timing. Teacher and student both use 5×; capacity reduction and magnification reduction must remain separate hypotheses.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Multi-Teacher Contrastive Distillation for Edge-Efficient Pathology Foundation ModelsTim Lenz, Maurice Heide, Marco Gustav, Nic G. Reitsam, and Jakob Nikolas Kather2026 · MICCAI Efficient Medical AI workshop (EMA) · peer-reviewedOfficial MICCAI satellite proceedings confirm the 2nd Efficient Medical AI workshop (EMA), not the main MICCAI track. LNCS volume 17265 is listed; DOI and pagination remain pending on the checked record. Full 11-page accepted manuscript inspected. Teachers and students use 2.0 µm/px (5×); this is capacity compression, not a demonstrated 20×→5× transfer method. Raspberry Pi timings concern individual encoder forward passes, not complete WSI processing.MuCoDi learns compact pathology encoders by multi-teacher contrastive distillation from fixed Virchow2, UNI2 and H-Optimus-1 tile embeddings. Its MuCoEdge students keep much of the teachers' downstream performance with small MobileOne or RepViT backbones. Crucially, the paper uses the same 5× sampling for teacher and student: it compresses model capacity, not a 20× evidence stream into a 5× one.Official workshop accepted manuscript · Author implementation and documented scope

Reading question: Which performance survives after reducing capacity, at the same physical sampling and downstream readout?

Evidence boundary: The 23 tasks are teacher-eligible; encoder timing excludes WSI IO and aggregation. Close average AUROC is not formal non-inferiority or cross-magnification validation.

Question-conditioned evidence acquisition across magnifications

PathPilot separates question-agnostic diagnostic localization from question-dependent evidence accumulation, movement, zoom and stopping. The agent accesses real multi-scale ROIs; it is neither low-only distillation nor a new MIL pooling operator. Its MIL experiment evaluates whether selected regions remain useful.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
PathPilot: An Adaptive Question-Conditioned Multi-Magnification Navigation Agent for Whole-Slide Image AnalysisZhengyang Xu, Han Li, Jingsong Liu, Ayako Ura, Shihui Zu, Shaohua Kevin Zhou, Nassir Navab, and Peter Schüffler2026 · arXiv · preprintarXiv:2610.16734v1, submitted 12 October 2026; manuscript says under review, with no independently verified acceptance. Full 24-page public manuscript inspected; referenced supplementary material was not attached or independently obtained. Code is promised on publication, not a verified release. Adaptive thumbnail/1.25×/2.5×/5×/10× exploration is not knowledge distillation or low-resolution-only inference. QA scores use a language-model judge; deterministic repeated runs do not establish population uncertainty.PathPilot separates where tissue looks diagnostically promising from what evidence a particular question still needs. CrossMagLoc learns question-agnostic region priorities across magnifications; EviRegionSearch uses a VLM, language-model memory and explicit evidence gaps to move, zoom or stop. This is adaptive acquisition of real multi-magnification evidence, not distillation into a low-resolution-only student.Inspected arXiv v1 manuscript

Reading question: What evidence is missing, which region and scale can supply it, and what justifies stopping?

Evidence boundary: Small morphology subsets and a language-model judge limit inference. A same-institution cohort called external is not independent-site validation; deterministic repeatability is not patient uncertainty.

Complete maintained paper route

Section titled “Complete maintained paper route”

51 canonical readings · 2009–2026· 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 51 readings in maintained order

  5. #52018Journal of Pathology InformaticsPeer reviewedL2–L3

    Implementing the DICOM Standard for Digital Pathology

    Markus D. Herrmann, David A. Clunie, Andriy Fedorov, Sean W. Doyle, Steven Pieper, Veronica Klepeis, Long P. Le, George L. Mutter, David S. Milstone, Thomas J. Schultz, Ron Kikinis, Gopal K. Kotecha, David H. Hwang, Katherine P. Andriole, A. John Iafrate, James A. Brink, Giles W. Boland, Keith J. Dreyer, Mark Michalski, Jeffrey A. Golden, David N. Louis, and Jochen K. Lennerz

    Official artifacts: Official DICOM pathology standard portal

    Metadata checked

  17. #172023IEEE 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

  26. #262025MICCAIPeer 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

  27. #272025arXivPreprintL3

    Magnification-Aware Distillation (MAD): A Self-Supervised Framework for Unified Representation Learning in Gigapixel Whole-Slide Images

    Mahmut S. Gokmen, Mitchell A. Klusty, Peter T. Nelson, Allison M. Neltner, Sen-Ching Samson Cheung, Thomas M. Pearce, David A. Gutman, Brittany N. Dugger, Devavrat S. Bisht, Margaret E. Flanagan, and V. K. Cody Bumgardner

    Preprint; arXiv comments report submission to the AMIA 2026 Informatics Summit, not acceptance.

    Official artifacts: Official model

    Metadata checked

  28. #282025Scientific 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

  33. #332026arXivPreprintL2–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

  34. #342026Medical 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

  39. #392026Engineering 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

  40. #402026Computers & 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

  41. #412026Medical 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

  42. #422026arXivPreprintL3

    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

  43. #432026ECCVAccepted · proceedings pendingL3

    Scaling Whole-Slide Pathology Foundation Model Pretraining with Billions Off-the-Shelf Tokens

    Honglin Li, Zhongyi Shui, Chenglu Zhu, and Lin Yang

    ECCV 2026 acceptance and the manuscript are verified from the official program. No Springer version of record or author-owned public code/checkpoint was verified. Section 4.1 explicitly combines HistAI with about 10,000 private WSIs; do not describe pretraining as public-only. PathTokScale and PathTokenScaling both occur in the manuscript.

    Official artifacts: Official ECCV 2026 Poster

    Metadata checked

  45. #452026arXivPreprintL3

    WILSON - a pathology foundation model framework for patient-level analysis and diagnostic text generation

    Saghir Alfasly, Wataru Uegami, Sobhan Hemati, Wenchao Han, Xiaojia Tang, Kevin Thompson, Daniel Stone, Ghazal Alabtah, Saba Yasir, Michael R. Lucas, Eric W. Klee, Cheryl L. Willman, Judy C. Boughey, Matthew P. Goetz, Krishna R. Kalari, and H. R. Tizhoosh

    arXiv v1 submitted 2026-09-20. No peer-reviewed record was verified. The manuscript says model release is planned after publication and institutional clearance; Mayo training WSIs/reports are not publicly downloadable.

    Official artifacts: Full manuscript and supplementary material

    Metadata checked

  46. #462026arXivPreprintL3

    Computationally Efficient Pathology Segmentation using Knowledge Distillation from Foundation Models

    Jiaqi Lv, Yijie Zhu, Saki Okada, Maxwell Stanley Renna, Robert Goldin, Stefan Antonowicz, and Shan E Ahmed Raza

    arXiv v1 (2026-09-03), submitted to Medical Image Analysis according to the authors, not accepted. Verified methods and selected results on 2026-10-02. TIAToolbox is the stated release destination; availability of these exact weights was not independently confirmed. Patch-throughput measurements are not end-to-end WSI latency.

    Official artifacts: Versioned manuscript · TIAToolbox — stated release destination

    Metadata checked

  47. #472026NeurIPSAccepted · proceedings pendingL3

    PathNavigate: A Training-Free Pathology Agent with Surprise-Guided Scan and Shared Slide Memory for Whole-Slide Image VQA

    Chunze Yang, Qidong Liu, Wenjie Zhao, Yue Tang, Jiusong Ge, Di Zhang, Jiashuai Liu, Lei Wu, Junbo Lu, Ni Zhang, Xian Wu, Zeyu Gao, and Chen Li

    Official NeurIPS 2026 poster 154224 verifies acceptance; program title shortens 'Whole-Slide Image VQA' to 'Whole-Slide VQA'. Review uses arXiv:2605.23559v1, not an uninspected camera-ready. Backbone-frozen navigation includes test-time memory optimization and high-magnification evidence; not 5×-only deployment.

    Official artifacts: Official NeurIPS 2026 program · Versioned full manuscript · Official inference implementation

    Metadata checked

  48. #482026NeurIPSAccepted · proceedings pendingL3

    Nonparametric Distribution Matching for Self-Supervised Whole-Slide Image Condensation

    Duong M. Nguyen, Trong Nghia Hoang, Hang Thi Nguyen, Thanh Trung Huynh, Phi Le Nguyen, and Minh N. Do

    Acceptance independently verified by NeurIPS 2026 poster 153235 and author repository. Review uses arXiv:2610.00678v1, submitted 2026-09-30. Repository contained only a README on 2026-10-08. Feature condensation at 20× is not low-magnification encoder distillation; Table 1 tests on full feature bags.

    Official artifacts: Official NeurIPS 2026 program · Versioned full manuscript · Author repository (README only at audit)

    Metadata checked

  49. #492026MICCAIPeer 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

  50. #502026MICCAI Efficient Medical AI workshop (EMA)Peer reviewedL3

    Multi-Teacher Contrastive Distillation for Edge-Efficient Pathology Foundation Models

    Tim Lenz, Maurice Heide, Marco Gustav, Nic G. Reitsam, and Jakob Nikolas Kather

    Official MICCAI satellite proceedings confirm the 2nd Efficient Medical AI workshop (EMA), not the main MICCAI track. LNCS volume 17265 is listed; DOI and pagination remain pending on the checked record. Full 11-page accepted manuscript inspected. Teachers and students use 2.0 µm/px (5×); this is capacity compression, not a demonstrated 20×→5× transfer method. Raspberry Pi timings concern individual encoder forward passes, not complete WSI processing.

    Official artifacts: Official workshop accepted manuscript · Author implementation and documented scope

    Metadata checked

  51. #512026arXivPreprintL3

    PathPilot: An Adaptive Question-Conditioned Multi-Magnification Navigation Agent for Whole-Slide Image Analysis

    Zhengyang Xu, Han Li, Jingsong Liu, Ayako Ura, Shihui Zu, Shaohua Kevin Zhou, Nassir Navab, and Peter Schüffler

    arXiv:2610.16734v1, submitted 12 October 2026; manuscript says under review, with no independently verified acceptance. Full 24-page public manuscript inspected; referenced supplementary material was not attached or independently obtained. Code is promised on publication, not a verified release. Adaptive thumbnail/1.25×/2.5×/5×/10× exploration is not knowledge distillation or low-resolution-only inference. QA scores use a language-model judge; deterministic repeated runs do not establish population uncertainty.

    Official artifacts: Inspected arXiv v1 manuscript

    Metadata checked