An Improved Skin Lesion Matching Scheme in Total Body Photography.

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Total body photography (TBP) is crucial for early malignant melanoma detection via temporal skin surface monitoring.
• Automated TBP image acquisition and processing were previously demonstrated, with skin lesion matching as a key component.
• Existing lesion matching algorithms require improvement to minimize false positive and negative outcomes.

Purpose of the Study:

• To enhance the accuracy of skin lesion matching algorithms used in automated TBP.
• To reduce false positive and negative identification rates in the detection of changes in pigmented skin lesions.
• To improve the reliability of automated melanoma monitoring systems.

Main Methods:

• Developed two novel lesion matching methods: 3D point cloud rigid transformations and region-specific nonrigid coordinate plane deformations.
• Implemented a robust outlier rejection procedure utilizing progressive graph matching for both methods.
• Created a diverse ground truth dataset from scanner images to specifically address false positive matching scenarios.

Main Results:

• The nonrigid deformation approach demonstrated superior performance compared to the rigid transformation method and the original algorithm.
• The novel methods achieved high precision and recall rates, with the nonrigid approach reaching 99.92% precision and 81.65% recall in complete interpositional matching.
• Significant improvements in reducing false positive and negative outcomes were observed, enhancing the accuracy of lesion identification.

Conclusions:

• The developed nonrigid deformation method represents a significant advancement in automated skin lesion matching for TBP.
• This improved accuracy enhances the potential for early and reliable detection of malignant melanoma.
• The findings support the advancement of automated systems for longitudinal skin monitoring and melanoma surveillance.