Advancements in Skin Cancer Diagnosis: A Literature Survey and Hybrid Approach Employing SVM and DNN Models with Results Analysis

This study combines a thorough literature review with the development of a novel framework for image analysis to present a comprehensive study in the field of skin cancer detection. Traditional machine learning, specifically Support Vector Machines (SVM), is combined with deep learning via a Deep Neural Network (DNN) in this methodology. The proposed framework starts with users loading skin cancer images via a user-friendly graphical interface, with the option of including ground truth images for reference. Preprocessing steps include grayscale conversion, Gaussian Low Pass Filter (GLPF) application, and segmentation via thresholding and region growing. When a ground truth image is provided, the segmented image is evaluated for accuracy. For further analysis, image features such as area, perimeter, color features, variance, and symmetry are extracted. The script prepares data for both SVM and DNN models during the training phase, using traditional feature extraction and deep learning on the Skin Cancer dataset. The trained SVM model and DNN are used to categorize features extracted from the input skin cancer image, with the predicted class (Benign or Malignant) displayed in the GUI. The graphical interface displays visualizations of the input image, ground truth, preprocessing results, and segmentation results, as well as quantitative metrics such as accuracy, sensitivity, and specificity for in-depth analysis. Initial experimental results show that the combined SVM and DNN approach is effective in skin cancer detection, laying the groundwork for future research and refinement in this critical domain.