Advancements in AI: Deep-Learning Models Mimic Pathologists in Biopsy Analysis

Researchers have developed innovative deep-learning models capable of analyzing tissue biopsies with a level of expertise comparable to seasoned human pathologists. Traditional AI training for medical image diagnosis often requires meticulous pixel-level annotations, a process demanding significant time and specialized effort from medical professionals. To address this challenge, a team from MedSight AI Research Lab, China Medical University, and the National Joint Engineering Research Center introduced a novel approach utilizing eye-tracking technology during pathological review.

This methodology involves recording eye movements, zooming, and panning behaviors of experienced pathologists as they examine whole-slide images of skin tissue samples. The collected data helps train the Pathology Expertise Acquisition Network (PEAN), a deep learning system that learns to decode the visual patterns and regions of interest identified by experts, thus capturing diagnostic expertise without extensive pixel annotation.

In a recent study published in Nature Communications, the team demonstrated that PEAN could accurately predict suspicious regions within tissue samples, achieving a classification accuracy of 96.3% and an AUC of 0.992 on known datasets. Its performance on new, unseen samples still yielded an impressive accuracy of 93% with an AUC of 0.984, surpassing other AI models. Moreover, when the regions highlighted by PEAN were used to train other diagnostic models, the overall accuracy and diagnostic confidence significantly improved.

This breakthrough holds promise for revolutionizing digital pathology by reducing the workload on expert pathologists, enhancing diagnostic precision, and enabling scalable training of AI models. The researchers envision expanding PEAN’s capabilities to support personalized diagnosis, lesion detection, and large-scale multimodal diagnostic systems, ultimately aiming to create digital replicas of expert pathologists that can operate efficiently across diverse clinical settings.

The approach highlights how integrating eye-tracking technology with deep learning can bridge the gap between human expertise and automated analysis, fostering advancements in medical diagnostics and AI-driven healthcare solutions.

Source: https://medicalxpress.com/news/2025-08-scientists-deep-scrutinize-biopsies-human.html