Computer aided diagnostic support system for skin cancer using ELM classifier

Melanoma cancer is the most fatal skin cancer that has been on rise in recent years. Early detection will increase the survival rate but it's challenging to detect. The severity of melanoma was usually diagnosed with invasive methods like biopsy. Alternatively, less time consuming and simplified methods were generated over time, yet a lot has to be improved. Researches proved that a computer aided automated skin cancer diagnosing method helps in processing the dermoscopic images in an efficient way. In this paper, image processing algorithms were used to decompose and fragment the original dermoscopic image and classify the lesion as melanoma or benign. In the proposed method, we used a set of composite feature vectors and an Extreme Learning Machine (ELM) classifier to detect the melanoma in dermoscopic images. Experimental analysis is done by calculating the total dermoscopic score using the feature extraction method. A set of composite features is extracted and is given as an input to the ELM classifier along with the total dermoscopic score (TDS). The sensitivity and specificity obtained using TDS score is 91.67% and 82.22% respectively and using composite features with TDS is 94.52% and 93.18% respectively. The generated receiver operating curve (ROC) clearly shows that the accuracy is increased to 94.01% and the area under the curve (AUC) 0.956 which is close towards 1, proves that the proposed method is more accurate and effective.