Designing a Retrieval-Based Diagnostic Aid using Effective Features to Classify Skin Lesion in Dermoscopic Images

Designing an automated system for the prior detection of melanoma in dermoscopic images can reduce the mortality rate, which has been raised due to skin cancer. Early diagnosis is thus pivotal to increase the survival rate and to prevent demolition caused due to cancer. The focus of this experiment is to develop a non-invasive digitized dermoscopic model to evaluate and diagnose the skin lesion. In this paper, SVM is used as the state-of-the-art approach for the classification of skin lesions as benign or malignant. PH2 dataset is employed in conducting the experiment. For better exploration and analysis of the image, after segmentation using thresholding method, feature extraction is thus performed using ABCD rule, which diagnoses the melanoma based on four visual criterions. The images are being analysed sequentially by the designed system for extracting the affected area. The decision is taken by the system based on characteristics which determine the degree of damage. The experimental results demonstrate the performance of the designed system by achieving the sensitivity 100%, specificity 87.5% and accuracy 92.5% on a PH2 data set, which is quite higher compared to the prior work.