Abstract:
A computer aided decision support system (CAD) has the tendency to improve the diagnosis of melanoma disease in skin using dermoscopy images. A CAD system provides quantitative and timely assessment as compared to the manual procedure. This research applies Machine Learning to develop a CAD system which will facilitate the physicians. A complete pattern recognition system has been developed which comprises of four vital stages to conform the analysis of skin lesions by the clinicians. These include pre-processing, segmentation, feature extraction and classification. The data-set contains images and annotations provided by physicians. Pre-processing cares for elements which degrade the overall performance. The main concern in this phase was to detect hair in dermoscopy images and remove them. Segmentation is an imperative preprocessing step for CAD system of skin lesions. Segmentation is performed using multi-thresholding technique in this study. Feature extraction of segmented skin lesions is a vital step for the implementation of accurate decision support systems. In the manual diagnosis process, the physicians are interested in examining a specific area of the dermoscopy image which is clinically significant region in a lesion. We have developed a versatile feature set where features come out of various different domains including histograms and Dominant Rotated Local Binary Patterns (DRLBP). Such a region is expected to have more information relevant to the detection process. The results obtained from classification using the proposed feature matrix showed superiority when compared with previous work done in this regard.
