Adaptive Attention-Based U-Transnet Architecture with Optimized Metaheuristic Classifier for Early Liver Tumor Detection

The work presented here gives a novel approach for detecting liver tumours from medical images.The proposed approach is the combination of the latest segmentation and classification techniques that results in early detection of liver tumours with higher accuracy.The methodology applies three phases, the first is the application of anisotropic diffusion filtering that enhances the image quality without disturbing the structural information.After filtering, the second phase makes use of the U-Net architecture and attention-based transformers (U-TransNet) segmentation model for precisely detecting the boundary delineation and tumour detection.The results show Intersection over Union (IoU) as 98%, and dice scores are 99.01%.The third phase in the proposed method applies a support vector machine optimised using a bio-inspired whale optimisation algorithm.The results measured using parameters like accuracy, recall, precision, and F1-score are close to 99%, considering class imbalance effectively in early stages.Comparative analysis in this study validated that the performance of this is better in high-end and advanced methods in segmentation as well as classification techniques.The proposed system has better performance over existing and therefore has a potential of accurate diagnostic of liver tumors for proper treatment of patients.