췌장 관 선암종의 디지털 병리이미지에서 AI 활용의 임상적의의

초록

Pancreatic Ductal Adenocarcinoma (PDAC) is the mostcommon and deadly form of pancreatic cancer. Currently,histopathological diagnosis and prognosis of PDAC aretime-consuming and labor-intensive for pathologists. Recent advances in pathological AI research aim toalleviate this. We accumulated training data, distinguishingPDAC areas in Whole Slide Images (WSIs) based onmedical findings. Using this data, we trained a deepconvolutional neural network for supervised learning toautomatically interpret PDAC areas. The AI modelachieved high Dice scores and, by visualizing thesegmentation results of the predicted histological images,validated that PDAC diagnosis and identification ofassociated regions are automatically possible, similar topathologists. Additionally, the AI model, which showedhigh specificity, suggests its potential as a co-pilot forhttp://doi.org/10.6109/jkiice.2024.28.1.106106pathological diagnosis and annotation.

키워드