Comprehensive Summary
Lui et al developed an AI deep learning model to aid the early diagnosis of mycosis fungoides (MF) and inflammatory skin diseases. The model used 1157 cases of MF and inflammatory disease, including 2452 clinical images, and 6550 dermoscopic images that correspond to clinical data. The authors trained an AI model based on RegNetY400MF to recognize patterns that differentiate MF and other diseases. Then they compared three groups to test the effectiveness of the model: dermatologists alone, the AI alone, and dermatologists using the AI for help. The results showed that the AI was more accurate than dermatologists. More specifically, when dermatologists used the AI, their accuracy improved from about 71.5% to 82.9%.
Outcomes and Implications
MF is the most common type of cutaneous T-cell lymphoma, and early-stage MF is hard to tell apart from other inflammatory diseases. Performing a biopsy is able to diagnose early-stage MF, but it is invasive. Other noninvasive information, like patient medical history and clinical images, would significantly help early diagnosis of MF, but there is a lack of tools to use and process this information. The AI model created by the authors can reliably help doctors identify early MF, significantly improving early diagnosis without the need for biopsy.