Skin tumor area extraction using an improved dynamic programming approach.

Area of Science:

• Dermatology and Medical Imaging
• Computational Pathology
• Image Analysis

Background:

• Accurate border description of pigmented skin lesions is vital for melanoma diagnosis using dermoscopy and the ABCD rule.
• Effective skin tumor area extraction (STAE) is essential for precise border analysis.
• Challenges in STAE include uneven illumination, artifacts, and fuzzy lesion borders.

Purpose of the Study:

• To present a novel STAE algorithm based on improved dynamic programming (IDP).
• To enhance the accuracy of tumor border segmentation in dermoscopy images.

Main Methods:

• The STAE algorithm involves color space transform (CIE L(*)a(*)b(*)), pre-processing, rough tumor area detection, and refinement.
• The dynamic programming algorithm is improved with a local cost function incorporating color and texture weights.
• The method utilizes the CIE L(*)a(*)b(*) color space for better alignment with dermatological perception.

Main Results:

• The proposed STAE method was evaluated on 100 dermoscopic images.
• Performance was benchmarked against state-of-the-art algorithms using dermatologist-drawn borders as ground truth.
• The IDP-based STAE achieved high accuracy with 96.64% sensitivity, 98.14% specificity, and 5.23% error probability.

Conclusions:

• The IDP-based STAE method is a highly effective solution for segmenting tumor borders in dermoscopy images.
• This technique demonstrates superior performance compared to existing state-of-the-art segmentation methods.
• The algorithm accurately extracts tumor borders, aiding in clinical diagnosis.