143 Automated Anatomical Mapping of Hand Eczema

Hand eczema (HE) is a highly prevalent chronic disease affecting 15% of the population. Clinical grading systems are time-intensive and require training, making overall coarse assessment default approach in practice. We present an image-based deep learning method to automatically evaluate HE lesion surface determine their anatomical repartition. This retrospective study was based on two datasets: (A) 312 pictures (B) 215 hand pictures. Eleven dermatologists annotated lesions student labeled with 37 sub-regions. Separate segmentation models (DLM) were trained for A B. Predictions from both DLMs merged used generate textual reports patient conditions. DLM performance evaluated using precision sensitivity 95% confidence interval. Intra-class correlation predicted experts’ annotations also evaluated. The achieved respectively 75% (64-82) 69% (55-81), while anatomy average 83% (80-85) 85% (82-88) same. intra-class 0.94 (0.90-0.96). An example report is: “Both hands show palmar side, namely 51% fingertips, 34% fingers, 27% palms 4% wrists.” Our can precise stratification lesions. It reproducible has no intra- inter-observer variability. could be performed remotely enable frequent follow-up teledermatology treatment.