AttnEffNet-B4: an attention-augmented EfficientNet-B4 framework with fourier transformation for robust multi-disease diagnosis.

Early and precise disease detection in healthcare plays a vital role for better treatment results, healthier life expectancy, and good quality of care. However, many conventional AI diagnostic models are designed with a narrow focus on a single disease identification. This would limit their applications in the current-day healthcare system, particularly in remote and rural areas where there is limited access to specialized healthcare services. For the automated diagnosis, the study proposed a unified advanced multi-disease classification system, namely Attention Augmented EfficientNet-B4 (AttnEffNet-B4), incorporating Fourier transformation, a stratified cross-validation framework, and transfer learning with an attention technique. Various pre-trained convolutional neural networks, such as EfficientNetB0, VGG-16, VGG-19, ResNet-50, and InceptionV3, are also used in the study to evaluate a variety of medical imaging modalities, including chest X-rays, MRI scans, CT scans, and skin lesion images. To improve classification accuracy across a range of disorders, such as neurological, respiratory, dermatological, and ocular conditions, advanced preprocessing and feature extraction techniques are used in the proposed AttnEffNet-B4 approach. Experimental results show that the proposed model outperforms the various pre-trained approaches, reaching a peak accuracy of 97.69% on the training set and 94.47% on the testing set, which comprises multiple diseases. To validate the effectiveness of the suggested AttnEffNet-B4 model, a comparative analysis is further done with several pre-trained CNN models, followed by the stratified k-fold cross-validation. This work incites sustainable development goals for good health and well-being by providing early detection of diseases, access to health care, and reducing diagnostic errors through AI-based multi-disease classification.