Harnessing the power of MAGRes-UNet with progressive cyclical convolutional neural networks for effective skin lesion segmentation in dermoscopic image analysis

Skin cancer surveillance and histological diagnosis are depending on skin lesion segmentation . Despite recent advances in deep learning , most existing models still struggle to capture multi-scale contextual information and maintain precise boundary definitions. To overcome these limitations, this paper proposes Progressive Cyclical Convolutional Neural Network with Multi-Attention Gated Residual U-Netfor Effective Skin Lesion Segmentation in Dermoscopic Image Analysis (SLS-PCCNN-MAGRes-UNet). Initially, the input dermoscopic image is gathered from ISIC 2020 database. During the preprocessing stage , the Time-Frequency Domain Polarization Filtering (TFDPF) is used to eliminate all artifact sources such as bubbles and hair from skin images. The preprocessed imageries are fed into the Progressive Cyclical Convolutional Neural Network with Multi-Attention Gated Residual U-Net (PCCNN-MAGRes-UNet) for segmenting the Region of interest (ROI). The SLS-PCCNN-MAGRes-UNet model introduces a cyclical refinement mechanism that allows the network to revisit and progressively enhance the segmentation outputs through recurrent feedback across convolutional stages. The data is collected from ISIC 2020 dataset is used. The proposed SLS-PCCNN-MAGRes-UNet technique attains 0.98 dice coefficient, 0.97 jaccard index, 99.2% accuracy performs better than existing models. The simulation outcomes validate effectiveness of the SLS-PCCNN-MAGRes-UNet method in addressing the key challenges in dermoscopic image segmentation .