Modified CNNs for Automated Binary Classification of Skin Lesions: Benign vs.Malignant

Skin cancer is a significant concern among various types of cancers that influences individuals. A deadly kind of skin cancer is malignant melanoma. The majority of skin cancers either spread to other body parts or are lethal if not found and treated quickly. Computer-aided diagnosis tools that can classify dermoscopic images have advanced in medical technology. This research paper focuses a method for the skin cancer identification using convolutional neural network, specifically using image data collected for melanoma malignant cells. 10540 images with benign and malignant lesions forms a dataset. 85% of it is going towards training, while the remaining 15% is utilized for testing. The images are resized and normalized additionally the training dataset is subjected to data augmentation to increase its diversity and enhance the model's generalization capability. The augmentation operations include random horizontal and vertical flipping, and random zooming with specified height and width factors. These operations create variations of the original images, contributing to a more robust training dataset. To categorize an image as benign or malignant, feature parameters are retrieved after using the image processing algorithms. Accuracy, ROC curve, and confusion matrix are the calculation measures with the help of which the proposed model attains an accuracy of 91.62 %. Hence, the research demonstrates a notable result in employing the MOCNN model in order to detect skin cancer.