DualAtt-ViTResNet With Fennec Fox Optimized Hyperparameters for Automated Osteoporosis Classification Using Medical Imaging

Resumo

Osteoporosis is a progressive bone disease which is necessary to reduce fracture risk and complications. This paper aims to establish an automated classification system for osteoporosis based on automated image processing of medical images, and to generate a sophisticated method of combining different processing techniques, such as pre-processing, data augmentation, feature extraction, and classification modelling using advanced Deep Learning (DL) methods. Preprocessing of images involves resizing and the use of a Gaussian filter to eliminate noise from the image and thus improve image quality. Augmentation of images occurs through the use of Style GAN to create synthetic images, reduce the effects of data imbalance and improve the generalization of a model trained on limited sample sizes. Feature extraction occurs through the use of DenseNet121 which is demonstrated to be very efficient in producing a large number of high-quality hierarchical features. The techniques for classification of the extracted features are based on a novel Dual-Attention Vision Transformer and ResNet architecture (DualAtt-ViTResNet). It provides strong local feature representation of ResNet and the ability to acquire contextual information at the global level through vision transformers and their dual attention mechanism. Finally, hyper-parameter optimization of the DualAtt-ViTResNet architecture is accomplished using the Fennec Fox Optimization Algorithm (FFOA) which is used to improve accuracy, speed, and robustness of the classification task. A detailed experimental evaluation of the proposed framework indicates that the automated osteoporosis classification framework outperformed the existing frameworks with respect to higher accuracy of 0.9744, precision of 0.9767, Recall of 0.9789 and F1-Score of 0.9777. The proposed system offers an efficient solution for automated osteoporosis detection, supporting clinical decision-making and early diagnosis in medical imaging applications.