Computer Vision-Based Medical Image Segmentation Using Hybrid CNN and Transformer Architectures

Abstract

In computer-aided diagnosis, disease monitoring, treatment planning, and precision healthcare, medical image segmentation is a crucial task that allows the identification of the anatomic structures and pathological regions in biomedical images. Traditional convolutional neural network (CNN)-based segmentation models have shown a high level of local feature extraction, but tend to have limited global contextual information and lack of long-range dependency modeling, which results into erroneous boundary demarcation and low segmentation accuracy under traditional medical imaging conditions. This study attempts to address these drawbacks by proposing a hybrid CNNTransformer framework, in which the capability of learning spatial features of CNN backbones is combined with the ability to learn the global context of transformer-based attention mechanisms to improve the medical image segmentation. The proposed architecture uses hierarchical local feature extraction with CNN encoder and transformer modules to extract semantic dependencies of long range and multi-scale contextual features, enhancing the robustness and accuracy of segmentation. The standard medical image segmentation dataset was used to evaluate the effectiveness of the proposed method through an experimental approach in which preprocessing and augmentation methods were implemented to enhance model generalization and efficiency in the training process. The proposed model was evaluated with the well-known segmentation measures, such as Dice Similarity Coefficient (DSC), Intersection over Union (IoU) and pixel-wise Accuracy. Experimental findings have shown that the hybrid framework achieves better segmentation performance than the conventional CNN-based frameworks because the framework provides better representation of the features, less false segmentation regions and accuracy in the boundaries. The suggested method demonstrated significant progress on Dice score, IoU, as well as the overall consistency of segmentation on difficult samples of medical imaging. The created framework provides strong clinical importance in that it enables more confident automated diagnosis, lessening manual annotation work, and enhances the decision making ability in intelligent health care system and computer-aided medical imaging software.