Hair detection and lesion segmentation in dermoscopic images using domain knowledge

Area of Science:

• Medical Image Analysis
• Computer-Aided Diagnosis
• Dermatology

Background:

• Automated segmentation and hair detection in dermoscopic images are crucial for computer-aided diagnosis (CAD) of melanocytic lesions.
• Existing methods face challenges due to artifacts, skin texture variations, and smooth lesion boundaries, impacting accuracy.

Purpose of the Study:

• To develop an automated algorithm for hair detection and lesion segmentation using lesion-specific properties to enhance accuracy.
• To improve the reliability and robustness of segmentation in CAD systems for melanocytic lesions.

Main Methods:

• A novel hair detection algorithm considering dermoscopic hair properties was designed.
• A novel chroma-based geometric deformable model with a chrominance-based speed function was employed for lesion segmentation.
• Automatic initialization of the contour and chrominance-based speed function ensured robust and flexible segmentation.

Main Results:

• The algorithm achieved high average accuracy (93.4% on PH2, 94.6% on ISBI 2016 datasets).
• Excellent sensitivity and specificity were recorded (e.g., 87.6% sensitivity, 95.3% specificity on PH2).
• Low overlap scores (11.52% on PH2, 7.20% on ISBI 2016) indicate precise segmentation.

Conclusions:

• The proposed approach demonstrates reliable performance for hair detection and lesion segmentation.
• The algorithm's accuracy and robustness make it suitable for integration into computer-aided diagnosis systems.
• This work contributes to advancing automated analysis of dermoscopic images for improved diagnostic capabilities.