Digital Pathology Podcast

Send us Fan Mail
Can you use AI to write scientific papers without fabricated references, journal trouble or learning to code? I recorded this episode at the airport on my way to ESTP 2026 in Kraków. It's my practice run of the talk I gave there.
When deep learning came into pathology, people asked me if I was going to learn to code. I decided no. Large language models kind of proved me right. In this episode I walk you through how I use LLMs for writing papers, responding to co-authors, building slides and even running code, and the four rules that keep it safe:
Domain expertise is the prompt
Supply the sources
Encode corrections
The subject matter expert verifies
In this episode:
What the LLM does vs. what the pathologist does, task by task
What Toxicologic Pathology, Veterinary Pathology and Elsevier journals require when you disclose AI use
Pathology LLM tools already working on tox study reports, image analysis and whole slide images
The reference I didn't check that almost cost me a publication
Why an LLM will happily justify switching patients from Tylenol to acetaminophen
Automation bias, model bias, and how to start with the tools your institution has approved
Going to PathVisions 2026 in San Diego, Oct 16-18? Find me at the Hamamatsu booth #201. I'm also recording 8 short podcast interviews on site, Saturday and Sunday. Share your digital pathology story, grab a slot: https://calendar.google.com/calendar/appointments/schedules/AcZssZ0yjFhaL0kGJ7bZ18oITwyw-qY2yxvcDEyfqpj3LBulrRdiO4fRUkgM0Avlj_oJkIaw-0ZjhaC9
Links
Free book, Digital Pathology 101: https://www.aleksandrazuraw.com/book
Related episode:
How to teach AI to healthcare professionals, with Dr. Candice Chu: 
Watch the video version on YouTube: https://youtu.be/eY7UywT-zRM?si=qK88eECJqoA-DqKF

References
Zuraw & Aeffner, Vet Pathol 2022. doi:10.1177/03009858211040484
Turner et al., Toxicol Pathol 2021. doi:10.1177/0192623321990375
Thirunavukarasu et al., Nat Med 2023. doi:10.1038/s41591-023-02448-8
Lu et al., J Transl Med 2026. doi:10.1186/s12967-026-08829-0
Doktorova et al., Arch Toxicol 2026. doi:10.1007/s00204-026-04414-y
Trost et al., Nat Med 2026. doi:10.1038/s41591-026-04357-y
Chen Y et al., Nat Cancer 2026. doi:10.1038/s43018-026-01220-4
Chu, Front Vet Sci 2024. doi:10.3389/fvets.2024.1395934
Jain et al., Cureus 2025. doi:10.7759/cureus.81618
Kurland et al., Neurosurgery 2025. doi:10.1227/neu.0000000000003354
Painter et al., JAMIA Open 2025. doi:10.1093/jamiaopen/ooaf003
Chen S et al., NPJ Digit Med 2025. doi:10.1038/s41746-025-02008-z
Dratsch et al., Radiology 2023. doi:10.1148/radiol.222176
Zack et al., Lancet Digit Health 2024. doi:10.1016/S2589-7500(23)00225-X
FDA Warning Letter 722591, Apr 2026
EMA/FDA, Guiding principles of good AI practice in drug development, Jan 2026
Huang & Chu, Front Vet Sci 2026. doi:10.3389/fvets.2026.1801756

Support the show
Get the "Digital Pathology 101" FREE E-book and join us!

Send us Fan Mail
What happens when AI models that perform almost perfectly on scattered cervical cells become less reliable than a coin flip on crowded cell groups?
In DigiPath Digest #51, I examine what this performance gap tells us about artificial intelligence in cytopathology.
The first paper evaluated six convolutional neural network models trained to distinguish benign from high-grade lesions using scattered cervical cytology cells. The models achieved AUCs ranging from 0.950 to 0.996 on the original images.
When the same models were applied to hyperchromatic crowded cell groups without retraining or threshold recalibration, their AUCs fell to 0.385-0.683. Different architectures also failed differently. Some overcalled benign clusters, while others missed high-grade clusters.
This isn’t simply a technical problem. It illustrates a practical rule for pathology AI: a model should only be trusted for the morphology, specimen type, imaging system, and intended use on which it has been directly validated.
I also review the current evidence for large language models in cytopathology. Potential applications include structured reporting, diagnostic support, quality control, education, research, and workflow integration.
Some early results appear promising, but each comes with important limitations. One structured-reporting application achieved 99.4% accuracy at a single institution. A diagnostic-support model included the correct answer among its top 10 differentials in 59.1% of general medicine cases. A hybrid quality-control system flagged 84% of errors associated with amended reports, but its false-positive rate wasn’t reported.
Most importantly, the review found no language model specifically trained and clinically validated on cytopathology reports.
The takeaway is straightforward: we’re still working with narrow AI. Strong performance in one setting doesn’t guarantee performance when the cells, preparation, scanner, institution, or clinical task changes.
Low-risk applications may offer the most practical starting point. Text extraction, completeness checks, report consistency review, and quality-control flagging could reduce repetitive work without asking an unvalidated model to make the final diagnosis.
Highlights with timestamps
00:00 - Welcome to DigiPath Digest #51 and the new lunch-and-learn time
01:40 - How two image models performed worse than a coin flip on cell clusters
02:29 - Image models, language models, and vision-language models
04:40 - Why AI adoption in cytopathology remains low
05:25 - Scattered single cells versus hyperchromatic crowded cell groups
08:14 - AUCs fall from 0.950-0.996 to 0.385-0.683
10:02 - How ResNet-50 and GoogLeNet failed differently
11:04 - What the attention maps revealed
13:22 - The intended-use lesson for pathology AI
17:02 - Current applications of large language models in cytopathology
18:39 - Structured reporting and the 99.4% accuracy result
19:32 - Diagnostic support, AMIE, and the top-10 limitation
21:09 - Quality-control applications and the missing false-positive rate
23:30 - Why cytopathology still needs domain-specific language models
24:54 - Retrieval-augmented generation, education, and research support
27:07 - Two AI families, one requirement: direct validation
28:35 - Context of use and intended-use validation
29:12 - Human-reviewed training data and destructive book scanning
32:02 - FDA and European approaches to evolving AI models
35:00 - Protecting patient and practitioner well-being
36:21 - Cytopathology-specific benchmarks and shared test sets
38:16 - Why low-risk AI applications should come first
39:50 - Cytopathology’s direct-to-digital advantage
40:39 - Digital Pathology 101 and Pathology Visions
Resources from this episode
Watch DigiPath Digest #51
Diagnostic performance of AI models trained on scattered single-cell images
Leveraging large language models to enhance cytopathology
FDA guidance on AI and context of use
FDA guidance on predetermined change control plans
Google Research: AMIE diagnostic medical AI
EU In Vitro Diagnostic Medical Devices Regulation
EU Artificial Intelligence Act
Digital Pathology 101
Pathology Visions 2026
Support the show
Get the "Digital Pathology 101" FREE E-book and join us!

Send us Fan Mail
What good is a powerful foundation model if it slows the pathologist down, can’t explain its result, or doesn’t fit the clinical workflow?
Foundation models are gaining attention across digital pathology. But they’re not finished clinical tools by themselves.
In this episode of the Digital Pathology Podcast, I speak with Panu Kauppila, Chief Product Officer at Aiforia, about what foundation models are, how they differ from traditional convolutional neural networks, and what it takes to make them useful for pathologists.
Panu describes a foundation model as a large, context-aware building block. To perform a specific pathology task - such as grading, segmentation, or mitotic counting - it must be combined with an adapter, a task-specific head, curated annotations, a usable interface, and integration with the laboratory workflow.
We also discuss one of the biggest practical constraints: speed.
A pathologist shouldn’t have to click a button and wait for an analysis. Panu explains why AI should run automatically in the background so the results are already available when the case reaches the pathologist’s worklist.
The conversation also examines the tradeoff between model size, computational cost, and clinical performance. Larger models may improve robustness and generalizability, but they can also require more processing power. For clinical applications, Aiforia focuses on smaller and medium-sized foundation models that provide the necessary quality without making the workflow slower or unnecessarily expensive.
Annotated data remains central. The foundation model supplies the underlying image understanding, while the task-specific head and controlled annotations determine how the model performs on a particular pathology problem. This structure also raises important questions about bias, data provenance, ownership, regulatory documentation, and explainability.
Finally, we look at multimodal AI models that combine pathology images with reports, genomic data, molecular information, and clinical outcomes. These tools could support more interactive, predictive, and prognostic applications - but only if they’re introduced through secure, controlled workflows with clear audit trails.

Episode Highlights
00:00 — Where does bias enter a foundation model workflow?
Panu distinguishes the underlying foundation model from the annotated dataset used to build the task-specific application.
00:27 — Meet Panu Kauppila
An introduction to Aiforia’s Chief Product Officer and the episode’s focus on foundation models in digital pathology.
01:05 — From radiology AI to digital pathology
Panu describes his background in medical device development, radiology, oncology solutions, and clinical AI implementation.
06:29 — Foundation models versus convolutional neural networks
What makes foundation models more context-aware, robust, and generalizable across image datasets.
07:07 — Image-only and multimodal foundation models
Why these two categories offer different capabilities and potential clinical uses.
07:50 — A foundation model is a platform, not a finished solution
The underlying model may understand image features, but it still needs a task, interface, and clinical workflow.
08:34 — Foundation model, adapter, and task-specific head
How these components work together to create an application for grading, mitotic counting, or another pathology task.
09:57 — The cost of larger models
Why increased robustness must be balanced against computational demands, inference speed, and affordability.
11:12 — Pathologists won’t wait for AI
Why even short delays can interrupt the clinical workflow.
11:46 — Running AI in the background
A workflow in which slides are scanned, analyzed automatically, and added to the worklist with results ready for review.
12:16 — Combining foundation models with curated annotations
How smaller task-specific datasets and adapter technology can produce practical pathology models.
15:24 — Generalizability across scanners, laboratories, and populations
How foundation models may make adaptation to new domains more manageable.
16:36 — Two datasets, two sources of potential bias
The regulatory questions created by an underlying foundation model and a separate controlled annotated dataset.
20:46 — Making foundation models accessible
Why a platform and user interface are necessary for pathologists and researchers who don’t work directly with code.
21:57 — Testing foundation models in Aiforia Create
How researchers can compare a CNN with supported foundation models in the same no-code environment.
25:45 — How foundation models are selected
Quality, licensing, model size, annotated data, and the requirements of the intended use.
26:57 — Training cost versus inference cost
Why a more expensive training iteration may still reduce total development costs if fewer iterations are needed.
32:23 — Pathologists are visual reviewers
The importance of segmentation quality and showing exactly what the model identified.
36:26 — The potential of multimodal AI
Combining pathology images with text, genomic information, molecular data, and clinical outcomes.
37:35 — Keeping multimodal AI inside a controlled environment
Privacy, security, regulatory oversight, and the risks of moving clinical information into consumer AI tools.
43:24 — Explainability in clinical pathology AI
Using semantic segmentation, object detection, instance segmentation, and annotated ground truth to show how results were calculated.
47:43 — Moving toward predictive and prognostic models
How established digital workflows could allow pathologists to contribute more information about likely outcomes.
49:45 — Research and clinical collaboration with Aiforia
How interested researchers and laboratories can connect through the Aiforia website.

Resources Mentioned

Aiforia
Aiforia Create no-code model-development environment
PathChat
Listen to the full discussion to understand what foundation models can add to digital pathology—and what still has to happen before they become practical, trusted clinical tools.
Support the show
Get the "Digital Pathology 101" FREE E-book and join us!

Send us Fan Mail
Does more experience automatically make a cytotechnologist more accurate—or does where they look first matter more?
In DigiPath Digest #50, I review a digital cytology eye-tracking study that challenges the assumption that diagnostic accuracy improves steadily with years of practice.
The researchers tracked the visual behavior of 100 board-certified cytotechnologists with 1 to 40 years of experience. They found no statistically significant linear relationship between years of experience and diagnostic accuracy. Instead, low-power field efficiency—the ability to identify an important target quickly within a wider field—emerged as the key predictor of high accuracy discussed in the study.
The study also examined whether this visual skill can be developed. Twenty-eight students completed an intensive three-month cytotechnology training program. After training, they located diagnostic targets more quickly and spent less attention on normal, nondiagnostic cells. In other words, they learned both where to look and what to disregard.
What could this mean for digital pathology education?
As AI-assisted workflows take on more of the exhaustive searching, cytotechnologists and pathologists may increasingly work as expert verifiers. That requires rapid target assessment, strong knowledge of normal morphology, and awareness of risks such as confirmation bias and cognitive fatigue.
The study has an important limitation: it used static images rather than dynamic whole slide imaging. The findings raise useful questions about visual expertise, training, and competency assessment, but they shouldn’t be generalized beyond the study design without further research.
Episode Highlights
00:00 – Welcome to DigiPath Digest #50 and introduction to the paper
04:10 – Why the traditional definition of professional expertise is changing
07:02 – Moving from exhaustive searching to verification in AI-assisted workflows
09:04 – How eye tracking was used with 100 board-certified professionals
10:25 – Years of experience versus diagnostic accuracy
13:13 – Experience-based caution and attention to sample information
15:16 – Low-power field efficiency as a predictor of high accuracy
17:05 – Practical low-power field demonstration using a whole slide image
20:15 – Searching versus detecting and the mental map of normal morphology
24:04 – Comparing high- and low-performer visual scan paths
25:27 – Cognitive filtering: knowing what not to examine
27:35 – Can visual efficiency be taught in three months?
29:55 – How AI may shift the human role from searcher to verifier
30:38 – Study limitations: static images versus dynamic whole slide imaging
32:37 – Could gaze efficiency influence future competency assessment?
33:44 – Digital pathology learning resources and closing thoughts
Resources Mentioned
Abstract and paper: Screening Efficiency Over Experience: Rapid Target Detection in Low-Power Field as a Modifiable Cognitive Biomarker for Diagnostic Accuracy in Digital Cytology
Listen to the full DigiPath Digest #50 recording to examine what the study found, what it didn’t prove, and how visual search skills could influence digital cytology training.
Support the show
Get the "Digital Pathology 101" FREE E-book and join us!

Send us Fan Mail
Can a treatment decision depend on whether one pathologist sees 45% biomarker positivity and another sees 55%?
Visual immunohistochemistry scoring helped establish precision oncology. But as targeted therapies become more sensitive to subtle biological differences, categorical scores such as 0, 1+, 2+, and 3+ may no longer capture the information needed to identify the right patients.
In this episode, I speak with three Roche experts:
Gordana Juric-Sekhar, MD, anatomic pathologist
Saleh Miri, PhD, Director of Digital Pathology AI Algorithms
Purvi Gaglani, Regulatory Affairs Lead for Digital Pathology
We discuss how computational pathology is changing companion diagnostics by moving biomarker assessment from visual estimates to continuous, cell-level measurements.
The conversation examines the limitations of manual IHC scoring, including interobserver variability, intraobserver variability, visual fatigue, borderline cases, tumor heterogeneity, and the inability of the human eye to measure complex spatial relationships.
Using TROP2 scoring in advanced non-small cell lung cancer as an example, Saleh explains the normalized membrane ratio. This computational metric measures protein expression at the cell membrane relative to total expression within the cell—something that can’t be reproduced through conventional visual scoring.
We also clarify the difference between computer-assisted scoring and a fully computational companion diagnostic. An assisted tool supports a pathologist’s visual interpretation. A computational CDx generates the biomarker measurement through algorithmic, cell-level analysis.
That doesn’t remove the pathologist.
Pathologists remain responsible for evaluating tissue quality, staining quality, scan quality, tumor selection, image analysis results, and the final clinical context. They can reject a stain, request a rescan, exclude inappropriate regions, question the result, or seek a second opinion.
The episode also examines the regulatory implications of computational companion diagnostics. Instead of evaluating a single IHC assay, regulators may need to assess the complete system - from tissue preparation and staining to scanning, image management, algorithmic analysis, display, and the final biomarker report.
Finally, we discuss what laboratories will need to implement these workflows, including validated scanning infrastructure, cybersecurity, tighter preanalytical process control, and training that helps pathologists interpret continuous computational measurements.

Episode Highlights

00:00 — Pathologists remain central to computational CDx
Why computational tools provide more precise measurements without replacing pathology expertise.
01:09 — Why companion diagnostics are changing
Visual IHC scoring helped launch precision oncology, but the model is approaching its limits.
04:53 — The current companion diagnostic landscape
How IHC, next-generation sequencing, liquid biopsy, and visual biomarker scoring are used today.
07:01 — The mathematical burden placed on the human eye
Why manually assessing tens of thousands of tumor cells requires pathologists to estimate rather than calculate.
08:39 — The borderline patient dilemma
A digital tool can distinguish measurements such as 74% and 76%, while that difference is difficult to reproduce visually.
09:27 — Why spatial context matters
Computational pathology can measure biomarker heterogeneity, clustering, and relationships between tumor and immune cells.
12:17 — Where manual scoring reaches its limits
Interobserver variability, intraobserver variability, fatigue, staining interpretation, and heterogeneous tumors.
18:29 — Moving from judgment calls to quantified measurements
Why the next stage of precision oncology requires information beyond human visual perception.
19:14 — Computer-assisted scoring versus computational CDx
The important distinction between helping a pathologist calculate an existing score and generating a new algorithmic measurement.
23:22 — Computational pathology and decentralized workflows
How digital images can support remote review, access to expertise, and second opinions.
27:48 — Why therapies require higher-resolution biomarkers
Modern targeted treatments may respond to biological differences that categorical scoring can’t capture.
32:09 — TROP2 in advanced non-small cell lung cancer
The episode’s example of a biomarker requiring computational measurement.
33:26 — Understanding the normalized membrane ratio
How the algorithm measures membrane expression relative to total protein expression at the individual-cell level.
35:46 — Working with regulators on a new diagnostic model
Purvi discusses global health authority engagement and the FDA Breakthrough Device Designation.
38:33 — The computational CDx as a system of systems
Why staining, scanning, image management, algorithms, displays, and reporting must be evaluated together.
40:11 — Changes to validated workflow components
How using a different scanner, monitor, or other component could fall outside the defined device configuration.
43:30 — Why computational pathology is becoming necessary
Continuous measurements can reveal biomarker-treatment relationships that may remain hidden within categorical scores.
49:12 — The pathologist’s role in the workflow
Reviewing sample, staining, scan, image, algorithmic analysis, and the final biomarker result.
53:39 — Digital second opinions
How image management systems can simplify collaboration without physically transporting glass slides.
56:43 — What laboratories need to prepare
Validated infrastructure, cybersecurity, preanalytical control, training, and digital pathology literacy.
58:49 — Learning to interpret computational results
The shift from visually estimated categories to continuous, quantitative biomarker measurements.

Resources Mentioned
Full discussion on YouTube: https://youtu.be/oKW1xC6TTZg
Listen to the full discussion to understand how computational pathology could change companion diagnostics—and what pathologists, laboratories, and regulators must prepare for next.

Support the show
Get the "Digital Pathology 101" FREE E-book and join us!