On July 1, Nan Tianhang, a doctoral student from the College of Medicine and Biological Information Engineering of NEU, and Zheng Song, a dermatologist from the First Affiliated Hospital of China Medical University, published the article titled "Deep learning quantifies pathologists' visual patterns for whole slide image diagnosis" in Nature Communications as co-first authors. This study was jointly completed by the College of Medicine and Biological Information Engineering of NEU and the First Affiliated Hospital of China Medical University over a period of five years. Qi Ruiqun (the First Affiliated Hospital of China Medical University), Gao Xinghua (the First Affiliated Hospital of China Medical University), and Cui Xiaoyu (NEU) are the co-corresponding authors, and NEU is the primary affiliation.
This study utilized eye-tracking technology to capture the visual behavioral characteristics of pathologists when reading films and designed a deep learning system named Pathology Expertise Acquisition Network (PEAN). This system, through its self-developed pathological image reading device (EasyPathology), extracts the professional knowledge of pathologists from their eye movement data and effectively transfers this knowledge to an intelligent model capable of autonomous diagnosis, thereby significantly reducing the reliance on expensive medical data annotation. The core advantage of PEAN lies in its ability to efficiently decode expert knowledge and accurately diagnose whole-slide images (WSI) while significantly reducing the annotation burden. The test results on over 6,000 whole-slide images containing five types of skin lesions show that PEAN has excellent diagnostic performance, with an AUC value as high as 99%, significantly surpassing traditional supervised learning and weakly supervised learning models.
The team led by Associate Professor Cui Xiaoyu from the College of Medicine and Biological Information Engineering, NEU has been conducting interdisciplinary research in the field of computational pathology for many years. The related achievements have been published in top international journals and conferences such as IEEE JBHI, Pattern Recognition, and ICLR. The Zhimou pathological image analysis system developed by the team has achieved the transformation of scientific research results and has been applied in more than 20 medical institutions across the country.
