Methodological review: Computerized analysis of pigmented skin lesions: A review
Artificial Intelligence in Medicine
| Hi-index | 0.00 |
The development of computerised systems which differentiate between suspicious tumours and clinically benign pigmented skin lesions can assist in improved early diagnosis of malignant melanoma, subsequently reducing mortality rates associated with this disease. One lesion feature indicative of malignancy is border irregularity. In this paper the theoretical fundamentals of a harmonic-wavelet based methodology for skin lesion border evaluation are described. Our methodology is applied to the boundaries of 30 cutaneous lesions and classification algorithms are consequently utilised to evaluate the effectiveness of descriptors for differentiating between regular / irregular borders and between benign / malignant tumours. Results indicate that generated parameters have high discriminative value when differentiating between benign and malignant lesions: maximum classification accuracy achieved was 93.3%, with 80% sensitivity.