Hybrid deep learning approach for early emphysema diagnosis combining fuzzy C-means, TransUNet, and faster mask R-CNN

Emphysema is a chronic condition of the lungs and it has to be diagnosed at an early age as it can harm the lungs effectively as they can be treated and controlled. The quality of the diagnosis depends on the efficient detection of the emphysematous areas in CT scans, yet the majority of the current methods fail in terms of noise, slight tissue differences, and delimiting the area composition. In this work, a new Hybrid IFCM-TransUNet-Faster Mask R-CNN framework to detect and localize prominently early emphysema cases will be proposed, featuring Improved Fuzzy C-Means (IFCM) preliminarily segmentation using noise-resistant Improved Fuzzy C-Means (IFCM) clustering, TransUNet preliminary feature extraction with transformers, and instance detection and localization provided by Faster Mask R-CNN. Such a combination is unique to offer pixel-level segmentation and object-level detection, as well as being more accurate and interpretable than single deep models. On 2000 CT images, the experiments reached a validation accuracy of 97.4 and a Dice coefficient of 0.97 and surpassed baselines on U-Net, Mask R-CNN, and DeepLabV3 + . The suggested system improves the minimization of false positives, boundary delineation improvement, and automated and understandable emphysema detection, which can pass clinical decision support.