Digital Pathology Podcast

228: GPT-5 and Gemini 2.5 Pro read pathology slides - here is how they did…

• Aleksandra Zuraw, DVM, PhD • Episode 228

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 24:15

Send us Fan Mail

I did something I've never done before for this episode — I went live from the middle of a national park. This is DigiPath Digest #42, broadcasting from the Great Sand Dunes National Park in Colorado via Starlink from my family road trip. Yes, it actually worked. And so did the papers.

This episode covers four papers that all ask the same uncomfortable question from different angles: how close is AI to being genuinely useful in real pathology practice — and what's still standing in the way? From LLMs interpreting cervical Pap smears, to AI guiding breast cancer treatment decisions from a simple H&E slide, to a practical roadmap for bringing generative AI into oncology workflows — this one covers a lot of ground.

I also introduced something new: my AI-powered paper summary podcast subscription. For $7 a month, AI hosts summarize digital pathology literature in a journal-club style so you can stay current without spending hours reading abstracts. I walk through how it works and why I built it.

What we cover:

  • [00:00] Going live from the wilderness — Starlink, sand dunes, and a very cold morning
  • [02:01] How I use AI-generated audio summaries to prep for each DigiPath Digest
  • [03:19] Paper 1: Can LLMs like ChatGPT and Gemini interpret cervical cytology? Spoiler: ~47–48% exact concordance — promising, but not there yet
  • [10:23] Bonus: My new AI-powered paper summary subscription — $7/month, journal-club style
  • [14:05] Paper 2: AI in oral oncology — CNNs for early lesion detection, multimodal prognostics, and the real barriers still blocking clinical adoption
  • [20:28] Paper 3: Generative AI in oncology — from chat tools to agentic EHR-integrated assistants, and why augmentation is the goal, not automation
  • [25:35] Paper 4: Computational pathology in breast cancer — predicting BRCA1/2, HER2, Oncotype DX, and treatment response from standard H&E slides
  • [31:39] Final thought: the floor just got raised for all of us — how I think about new technology in pathology

Resources & Links:

Support the show

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


00:00:01 | Aleks | Welcome, my trailblazers. We have not done that before, and I will need a bunch of confirmation from you that this is actually working because we are broadcasting from the wilderness.

00:00:30 | Aleks | Let me know in the comments where are you tuning in from today because I bet you have not seen this version of me yet. I am tuning in from the Great Dunes National Park US. It's in Colorado.

00:00:49 | Aleks | And it's so freaking cold, but the sun is coming out. Let me know that you hear me because I don't know if my assistant can paste the picture in the comment of how I look with all my gear here.

00:01:17 | Aleks | So, I am traveling with my family, and we hear my kids in the background. I hope they don't come here and disturb me during the livestream. We have the Starlink Internet on the car, and it's actually working.

00:02:01 | Aleks | So, let's start on Starlink. The first paper. We have such cool papers today. I read the abstracts, and I listened to the audio versions, summarized by the AI that I publish on my new subscription feed of the podcast.

00:03:19 | Aleks | So can large language models like ChatGPT and Gemini interpret cervical cytology? LLMs have shown promise in medical imaging, but their utility in cytology remains underexplored.

00:04:23 | Aleks | They did ChatGPT and Gemini 2.5 Pro. Digital cervical pap test images of 100 cases were obtained from the Hologic education site. Hologic diagnosis considered the gold standard.

00:05:41 | Aleks | They were uploaded into ChatGPT and Gemini 2.5 Pro and prompted to provide a diagnosis based on the third edition of the Bethesda system for reporting cervical cytopathology.

00:06:37 | Aleks | Not that fantastic. Concordance for both LLMs for exact diagnosis: ChatGPT 47%, Gemini 48%. This increased to 66% for clinical grouping. GPT performed best for low-grade squamous intraepithelial lesions at 75%. Gemini showed highest concordance for high-grade lesions at 82%.

00:07:47 | Aleks | For sensitivity for detecting abnormal cytology: Gemini outperformed GPT — 84% sensitivity, 74% specificity. GPT identified glandular lesions and Gemini identified organisms more likely — 71% vs GPT's 20%. Current LLMs demonstrate ability to identify cytologic abnormalities but are not yet reliable independent cervical pap test interpreters.

00:10:23 | Aleks | I promised you that I'm gonna show the code for the AI summaries. This is a subscription of AI-powered paper summaries. It's just $7 a month. After two months, you will start seeing trends in what people are publishing.

00:14:05 | Aleks | Artificial intelligence in oral oncology: current advances and future potential in diagnosis, prognosis, and therapeutic decision making. Oral squamous cell carcinoma is a significant global health challenge because of rising incidence, late stage diagnosis, and poor five-year survival rates.

00:15:56 | Aleks | Diagnostic advancements: CNN-based models for image analysis, digital pathology, and mobile screening tools demonstrate high accuracy in early lesion detection and tumor grading. Prognostic tools leverage multimodal data including histopathology, radiomics, and genomic profiles to improve risk stratification and survival prediction.

00:17:24 | Aleks | Therapeutic planning assists in precision radiotherapy planning and surgical navigation, and predictive modeling of treatment responses through radiogenomic integration. Future directions emphasize federated learning, explainable AI, and multi-omics integration.

00:20:28 | Aleks | Generative AI is entering oncology. Large language models are the near-term workhorse because oncology runs on narrative text and structured data. Concrete uses include molecular tumor board synthesis, synoptic radiology and pathology drafting, and clinical trial matching.

00:21:40 | Aleks | Constraints include fragmented IT, price and provenance requirements, and hallucinations. Evaluation must move beyond leaderboards toward multicenter prospective designs reflecting clinical utility.

00:24:24 | Aleks | And what they concluded with: it's supposed to be augmentation, not automation. You're not automating doctors. You're augmenting them, because with their experience and intelligence, they can take it to the next level for patients.

00:25:35 | Aleks | How artificial intelligence applied to digital pathology could guide treatment personalization in breast cancer. The integration of AI with digital pathology has created computational pathology, which is valuable for diagnostic, prognostic, and predictive applications.

00:27:20 | Aleks | Several AI tools enhance performance compared to visual evaluation by pathologists for identification of invasive breast cancers, sentinel lymph node metastasis, histological grade, Ki-67, estrogen receptor, HER2, and PDL-1. Some are even cleared by the FDA.

00:28:32 | Aleks | There's strong potential in predicting from a simple H&E digital slide treatment response for neoadjuvant chemotherapy, anti-HER2 therapy, and endocrine therapy. Also predicting BRCA1/2, homologous recombination deficiency, and genomic signatures like Oncotype DX from H&E slides.

00:31:39 | Aleks | People are always afraid of new technologies — cars, computers. How I see these technologies is that the floor just got raised for all of us. Let's figure out how to create the next benefit for science, for patients.

00:32:48 | Aleks | Thank you so much for joining me today. I think the next episode can still be from the wilderness. Stay tuned. Subscribe to the list because then you're gonna get notification emails about what is happening in our digital pathology trailblazer community.