Skin Lesion Detection Algorithms in Whole Body Images

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Melanoma is a dangerous cancer requiring early detection for effective treatment.
• Automated detection systems analyze optical images of moles to identify neoplastic skin lesions.
• Full-body systems offer comprehensive patient analysis through multiple photographs.

Purpose of the Study:

• To present a prototype full-body system for melanoma detection.
• To assess the effectiveness of lesion detection and segmentation algorithms.
• To compare the accuracy of geometric parameter estimation for segmented lesions.

Main Methods:

• Developed a prototype full-body imaging system.
• Analyzed three lesion detection algorithms: deep learning, local brightness distribution, and correlation method.
• Evaluated algorithm fusion for improved detection sensitivity and precision.
• Calculated and compared geometric parameters of segmented lesions.

Main Results:

• Algorithm fusion achieved a detection sensitivity of 0.95 and precision of 0.94.
• High accuracy in estimating geometric parameters, with area estimation error below 10%.
• Accurate parameter estimation was particularly noted for lesions exceeding 3 mm in size.

Conclusions:

• The developed full-body system prototype demonstrates high accuracy in detecting and segmenting neoplastic skin lesions.
• The fusion of multiple detection algorithms enhances diagnostic performance.
• The system shows promise for improving early melanoma detection and patient outcomes.