A clinical decision support system for skin cancer classification using fractional gooseneck barnacle-enabled ensemble classifier.

Skin cancer is a prevalent disease worldwide, and among them, melanoma is the deadliest subtype. A lack of awareness regarding medical signs of skin cancer often leads to delayed diagnosis. Despite advancements in diagnostic technologies, skin cancer classification still faces significant challenges, and this obstructs early and accurate diagnosis. The challenges include misdiagnosis caused by symptom overlap with benign conditions, limited access to specialized dermatological care and high financial costs of screening procedures. In this work, a Fractional Gooseneck Barnacle Optimization-enabled Ensemble classifier (FGBO_Ensemble Classifier) is devised for classifying cancer in skin. At first, the input image is taken and fed to the image denoising phase, which is accomplished by the wavelet transform. Then, the Tversky Loss Function-based Comprehensive Attention Convolutional Neural Network (Tversky_CA-Net) performs skin lesion segmentation. Moreover, flipping and cropping are done in the image augmentation phase, and feature extraction is accomplished. At last, classification of skin cancer is done by an ensemble classifier that is formed by combining SpinalNet, Quantum Dilated Convolutional Neural Network (QDCNN) and Deep Kronecker Network (DKN). The ensemble classifier is tuned by FGBO, which is formed by integrating the Gooseneck Barnacle Optimization algorithm (GBO) and Fractional Calculus (FC). The effectiveness of the proposed FGBO_Ensemble Classifier is analyzed based on metrics, including accuracy, Negative Predictive Value (NPV), Positive Predictive Value (PPV), sensitivity, specificity, as well as F1-score with values of 92.82 %, 89.94 %, 91.71 %, 93.54 %, 92.43 %, and 92.61 %.