Development and Validation of an Artificial Intelligence-Based Image Classification Method for Pathological Diagnosis in Patients With Extramammary Paget's Disease.

Extramammary Paget's disease (EMPD) is a rare, malignant cutaneous adenocarcinoma with a high recurrence rate after surgical resection. Early diagnosis of EMPD is critical as 15%-40% of cases progress into an invasive form and resulting in a dismal prognosis. However, EMPD can be a diagnostic challenge to pathologists, especially in the grassroots hospital, because of its low incidence and nonspecific clinical presentation. Although AI-enabled computer-aided diagnosis solutions have been extensively used in dermatological pathological image analysis to diagnose common skin cancers such as melanoma and basal cell carcinoma, these techniques have yet been applied to diagnose EMPD. Here, we developed and verified a deep learning method with five different deep convolutional neural networks, named ResNet34, ResNet50, MobileNetV2, GoogLeNet, and VGG16, in Asian EMPD pathological image screening to distinguish between Paget's and normal cells. We further demonstrated that the results of the proposed method are quantitative, fast, and repeatable by a retrospective single-center study. The ResNet34 model achieved the best performance with an accuracy of 95.522% in pathological images collected at a magnification of ×40. We envision this method can potentially empower grassroots pathologists' efficiency and accuracy as well as to ultimately provide better patient care.