Implementing Automatic ABCD Rule for the Classification of Benign and Malignant Skin Lesions

Cases of melanoma have been gradually increasing for many years. 14,61,427 incidences of cancer were reported to be the projected total in India for 2022. The likelihood of developing cancer in India is one in nine persons where the North region of India has the highest AAR (Age adjusted rates) per 100,000 cases of cutaneous melanoma for both males and females, with 1.62 and 1.21, respectively. Males were more likely to develop non-melanoma of the skin or other skin malignancies in the East area than in the Northeast, where the incidence was greatest at 6.2 per 100,000. The most advantageous line of therapy is an early surgical excision. After removal at the earliest in-situ stage of melanoma, the life expectancy is unaffected. Early detection may be facilitated by automatic lesion analysis. Asymmetry, Border irregularity, Colour, and Diameter make up the ABCD rule of dermoscopy, which dermatologists use to quantify the findings and successfully identify malignant melanoma from benign lesions. Automatic ABCD trait recognition and differentiation between benign and malignant tumours enable early melanoma detection. The automated detection of hair and lesion borders is made possible by the employment of and geodesic active contours during the pre-processing stage. By modifying the kernel weights and attempting to increase the current accuracy, we hope to create a robust classifier model with OpenCV, Python, TensorFlow and UiPath. For the classifier model constructed, the obtained accuracy is 85.33% and other metrics such as precision of 84.48%, sensitivity of 85.33% and a specificity of 97.17% were obtained. When compared to other models that are already in use, this provides better results in terms of accuracy in predicting the rightful class of the lesion.