Pipelined Structure in the Classification of Skin Lesions Based on Alexnet CNN and SVM Model With Bi-Sectional Texture Features

The classification of skin lesions is crucial because it increases the likelihood that malignant skin lesions will be discovered early on, allowing for more effective treatment. Due to the abundance of lesion images and the possibility of human error, early detection can be difficult for dermatologists. This work aims to classify skin lesions using two pipelines that were designed using support vector machine (SVM) and AlexNet convolutional neural network (CNN) models. Pipeline-1 uses the AlexNet CNN, while pipeline-2 proposes a bisectional feature extraction approach with an SVM model. The skin lesion images are initially preprocessed and the lesion regions are segmented. The lesion regions are further subdivided into four regions based on the intensity mapping function. The bisectional features are then extracted from the subdivided regions and the extracted features are trained with the SVM model. The dataset used in the experiment is the HAM-10000 dataset and the PAD-UFES-20 dataset, which consists of dermatoscopic skin lesions images. Based on the models’ accuracy, sensitivity, DCI, specificity, and F1-score, the experiment’s findings will be assessed for five different skin lesion conditions. By accurately and effectively classifying skin lesions, the study’s findings will help in the diagnosis and treatment of skin disorders. The SVM pipeline performs better than the AlexNet CNN pipeline where the SVM pipeline and AlexNet CNN pipeline result in an accuracy of 98.66% and 97.68% respectively for the HAM-10000 dataset. The AlexNet CNN and SVM pipeline structure results in an accuracy of 96.87% and 98.10% respectively for the PAD-UFES-20 dataset.