Detection of multiclass non-melanoma skin cancer with multi-variable DCNN with hybrid gradient boosting optimizer

Skin cancer is a common type of cancer that emerges from the epidermis of the skin and spreads to other parts of the body. Among the types of skin cancer, non-melanoma skin cancer has fewer detection techniques and shows an increased case rate annually. Detecting the non-melanoma skin cancer remains a critical task as the existing methods struggle with the positional detection, potential of the tissues, and transparency and interpretability. In addition, the previous technologies faced difficulty with the large and poor-quality image dataset. To overcome these limitations, the proposed research designed a Multivariable Deep Convolutional Neural Network model for the accurate multiclass non-melanoma skin cancer. The model incorporates the Hybrid gradient boosting and the Hybrid attention module, which reduces the complexity and enhances the performance with accurate outcomes. The feature extraction process of the model includes two pre-trained models that are fused to achieve the best feature extraction, along with the shape and texture features. The ultimate goal of the model is to detect multiclass non-melanoma skin cancer with greater evaluation values, which achieves an accuracy of 96.8%, sensitivity of 97.56%, specificity of 96.03%, precision of 96.71% and F1-score of 97.13% with the Histopathological non-melanoma skin cancer segmentation dataset.