Simpler, faster, more accurate melanocytic lesion segmentation through MEDS

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Accurate segmentation of melanocytic lesions is crucial for early skin cancer detection.
• Existing segmentation techniques often lack the speed and accuracy required for real-world clinical applications.

Purpose of the Study:

• To introduce and evaluate Mimicking Expert Dermatologists' Segmentations (MEDS), a novel technique for melanocytic lesion segmentation.
• To assess MEDS for its accuracy, speed, and robustness compared to state-of-the-art methods and expert dermatologists.

Main Methods:

• MEDS employs a thresholding scheme that replicates dermatologists' cognitive segmentation processes.
• The technique incorporates optimizations for enhanced performance and efficiency.
• A single parameter controls the segmentation "tightness" for user-defined precision.

Main Results:

• MEDS achieves segmentation speeds significantly faster than current state-of-the-art methods, even on mobile devices.
• The accuracy of MEDS is comparable to or exceeds that of experienced dermatologists.
• Expert dermatologists exhibit less disagreement with MEDS segmentations than with those from other automated techniques or less experienced clinicians.

Conclusions:

• MEDS represents a significant advancement in automated melanocytic lesion segmentation.
• Its high accuracy and speed make it a promising tool for clinical use and potentially for mobile health applications.
• MEDS demonstrates the potential of mimicking expert cognitive processes for developing advanced medical image analysis tools.