New technological advancements and AI-supported workflows are making their way into the pathology lab to support clinicians – but can this technology be utilized in training? This is what a set of researchers across the US aimed to find out, exploring the possibilities of AI-assisted histopathology as a training tool for residents. We connected with lead researcher Min Zhu to learn more about this initiative.
What inspired this study, and why did you choose to focus on prostate biopsies?
Machine learning is changing pathology, with AI becoming a valuable tool for screening, diagnosis, and predicting outcomes. This led us to ask: could it also help train early-career pathologists? A learning tool based on AI could be especially useful for residents as they navigate the early, challenging stages of their careers.
Prostate pathology has been a pioneer in AI histopathology development, driven by strong demand stemming not only from a heavy workload but also from the need to address significant interobserver variability. While prostate pathology is the focus of the initial phase of our trial to develop an educational tool, it will pave the way for future development in other areas such as educational tools for GI pathology.
Please tell us about your study and its findings.
We recruited 15 pathology residents from different programs across the US – five each from PGY-1, PGY-2, and PGY-3 levels.
Because the project involved several teams – such as slide preparation, attending pathologists providing ground truth, AI development, IT support, and resident training – coordination was a major challenge. For each study round, residents had to complete their independent reads before accessing AI-assisted reads, which sometimes created scheduling issues. To solve this, the IT and AI teams built an individual on/off AI switch to simplify access.
We used the FlexLIS platform, developed by NoinoAI, which combines an easy-to-use image management system, case sharing, customizable AI annotations, and a built-in pathology ChatGPT Q&A. These features made it a strong tool for both independent and guided learning in pathology training.
Our results showed that AI-assistance can increase the residents’ reading accuracy in diagnosis and Gleason score grading. There was also a notable difference in outcomes between junior and senior residents: AI assistance had a greater impact on less experienced residents, helping them perform at a level closer to that of their more experienced peers.
Could this approach work for educating early-career pathologists or supporting experienced pathologists?
AI tools can help both new and experienced pathologists. One study found that AI improved the detection of lymph node metastases in bladder cancer for both recent graduates and those with over ten years of experience. Another study showed that AI helped general pathologists diagnose and grade prostate cancer at a level closer to that of GU specialists. Since pathology is a field of lifelong learning, AI has the potential to support ongoing education and professional growth throughout a pathologist’s career.
How do you see AI changing the landscape of histopathology education in the next five years?
I expect to see great change. First, tools like ChatGPT may replace traditional textbooks by offering portable, constantly updated information. Second, image-to-text technology can help residents when they face tough cases. Third, AI can suggest possible diagnoses and tests, which are key parts of training. It can also help residents preview slides and draft reports, making learning more efficient.
Is there a risk of residents becoming overly reliant on AI?
Machine learning is changing pathology much like smartphones changed traditional cell phones. It’s important to embrace this change with a positive mindset – using AI to boost our skills, not replace them. Teaching residents how to use AI wisely is a key part of this shift.
A strong curriculum that encourages independent thinking and solid diagnostic skills is essential. The goal is not for AI to replace critical thinking, but to amplify and support it.
It’s also important to help residents understand AI’s limits – what it can and can’t do – and to encourage them to build skills that go beyond AI, such as engaging in academic research and contributing to the development of improved AI tools.
What are your next steps for this research?
Our pathologists, AI team, and IT team at NovinoAI are working together to make FlexLIS a more complete educational tool. This includes adding features like image-to-text technology, grossing guides, reporting tools, and support for differential diagnosis.
We started with prostate pathology, but we’re now expanding to other areas like GI, GYN, cytology, and more.
What would you say to educators or program directors who are sceptical about integrating AI into medical education?
I’d say: AI isn’t a threat to traditional learning – it’s a chance to improve it.
These tools are meant to support, not replace, clinical reasoning. They can track resident progress, give real-time feedback, offer case-specific tips, and tailor learning to each individual. This helps residents grow and better meet ACGME standards.
Newsletters
Receive the latest pathology news, personalities, education, and career development – weekly to your inbox.
