DK-FocalNeXt: Deep Kronecker based FocalNeXt for skin cancer classification using dermoscopic images

The global burden of skin cancer continues to rise, posing a serious public health challenge. Accurate recognition is vital to ensure effective therapy and increase survival chances, even though classical diagnostic methods depend largely on the availability and experience of dermatologists. In many regions, especially those with limited medical infrastructure, it can lead to late diagnosis and negative impacts on patient health. Consequently, there is increasing interest in creating automated, precise, and accessible diagnostic tools to aid clinicians and facilitate early diagnosis of skin cancer. To moderate these complications, this article proposes an innovative approach named Deep Kronecker based FocalNeXt (DK-FocalNeXt) for the classification of skin cancer. The proposed framework initially takes skin images from the datasets as input. Then, image denoising is done using the Rudin-Osher-Fatemi model (ROF). The next stage involves segmenting the skin lesions using DAMMD-Net. Moreover, LogicMix is used for image augmentation, which promotes skin cancer classification by enhancing the diversity and size of training datasets. Thereafter, feature extraction was done to gain meaningful and distinctive representations of the lesions, which were essential for accurate classification. In the final stage, the skin lesions are classified into their respective categories using the DK-FocalNeXt, which is developed by integrating Deep Kronecker Neural Network (DKN) and FocalNeXt. Furthermore, DK-FocalNeXt has gained an accuracy of 96.61%, a weighted average precision of 95.69%, and a macro average recall of 95.38%.