<p>The identification of automated medical images is one of the most crucial assignments to be accomplished in the medical sector, making it possible to detect accurately the anatomical structures of a patient as well as the benign/malignant regions. Conventional deep learning models cannot handle low contrast, the presence of noise, and complex structural variations in the medical images. To overcome these drawbacks, a novel hybrid deep learning model is proposed that combines a convolutional neural network (CNN) equipped with a transformer module (TM) and generative adversarial networks (GANs) to improve segmentation of the images taken from several bio-medical datasets. Initially, images are pre-processed, normalized, and resized to maintain consistency even under different datasets. CNN performs feature extraction from the images and leads to the generation of feature maps. These maps are processed with the help of a self-attention-based transformer module for capturing long-range dependencies. GANs subsequently enhance the data and refine the segmentation results. The results are evaluated on different medical image datasets, which indicate that the proposed hybrid model outperforms existing segmentation models in terms of Dice Similarity Coefficients (DSC), Intersection over Union (IoU), and lower Hausdorff distances.</p>