Revisiting lesion tracking in 3D total body photography

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Melanoma, the deadliest skin cancer, requires early detection through monitoring nevi and new lesions.
• Existing 3D total body photography methods face challenges in accurate lesion pairing, noise sensitivity, and limited large-scale datasets.

Purpose of the Study:

• To develop a robust framework for longitudinal skin lesion tracking using 3D total body photography.
• To address challenges in lesion pair matching accuracy and dataset limitations for melanoma detection.

Main Methods:

• A framework utilizing correspondence maps and flow fields to align 3D textured meshes from different scans.
• Refinement of correspondence maps through advection along a constructed flow field for precise lesion matching.
• Development of the first large-scale dataset for skin lesion tracking, comprising 25K lesion pairs from 198 subjects.

Main Results:

• The proposed method achieves a 90.1% success rate for matching annotated lesion pairs (10mm criterion).
• A high matching accuracy of 98.1% was observed for subjects with over 200 lesions.
• Introduction of a novel, large-scale dataset facilitating advancements in skin lesion tracking research.

Conclusions:

• The developed framework significantly improves lesion matching accuracy in longitudinal skin lesion monitoring.
• The new large-scale dataset provides a valuable resource for training and evaluating skin lesion tracking algorithms.
• This work advances early melanoma detection capabilities through enhanced analysis of 3D total body images.