AN EFFICIENT MODEL FOR CLASSIFYING THE SKIN LESIONS USING HYBRID CONVOLUTION BASED MULTISCALE MOBILENETV3 WITH ADVANCED SEGMENTATION TECHNIQUE

Skin cancer is a type of cancer that affects the outermost layer of the skin. Enhancing the accuracy of diagnosis and the early detection rate is a crucial task in this context. Dermoscopy is a widely used technique for skin imaging, aimed at improving diagnostic precision and reducing skin cancer mortality. The challenge lies in the initial stages of skin cancer, where different subtypes can appear quite similar, leading to delayed detection. Dermatologists typically rely on visual assessment of dermoscopic images, which requires significant expertise, time, and can be influenced by individual biases. This is largely due to the fact that both cancerous and benign moles, as well as normal skin, can bear a striking resemblance, making accurate diagnosis challenging. However, patients and dermatologists are hesitant to utilize such CAD programs since the computation cycle behind model learning and feature coding is not widely recognized. Without a reasonable explanation, the approach to deep learning is a barrier to doctors making informed decisions. A dermatologist’s work on skin cancer research is aided by an automatic segmentation of the skin lesions system, which also serves as an infrastructure for early cancer detection. Over the years, various deep learning algorithms have been investigated for segmenting skin lesions. Nevertheless, training these deep models typically requires a substantial dataset with well-annotated examples. In the medical field, creating datasets is challenging due to the burden of documentation. Furthermore, when dealing with limited data, the risk of overfitting in neural networks significantly rises. So, an efficient skin lesion classification model using the deep learning method is developed in this paper. The demographic images are obtained from standard data sources. The obtained images are provided as the input to the Mobilenetv3 based TransUnet[Formula: see text] (MTUnet[Formula: see text]) for the segmentation process. The segmented images are now given to the classification stage. In the classification stage, the Hybrid Convolution-based Multiscale Mobilenetv3 (HCMMv3) is utilized to classify the segmented skin lesion image. The final classified skin lesion images are obtained from the HCMMV3 model. The accuracy of skin lesion classification is determined by carrying out extensive experiments on the developed model. The developed skin lesion classification method attains better and more reliable outcomes to enhance the overall process. Throughout the simulation analysis, the designed method is enhanced through 7.2% of CNN, 5.2% of ResNet, 6.86% of AlexNet, and 2.5% of MS-MobileNet in terms of accuracy.