Generative Artificial Intelligence-Driven Augmentation With Deep Feature Fusion Model for Rare Disease Detection and Classification

Rare disease detection focuses on identifying rare health conditions that affect a tiny fraction of people. Detection is challenging because symptoms are often ambiguous and information is limited. In recent years, retinal diseases have become a serious problem worldwide. Progressing a computer-assisted diagnosis (CAD) method is highly beneficial and functional for ophthalmologists. Proper detection and treatment can save eye vision. Deep learning (DL) has advanced across various scientific fields and is leveraged for numerous healthcare tasks. This research presents a Generative Artificial Intelligence-Assisted Adaptive Feature Fusion Network for Rare Retinal Disease Detection (GAIAFF-RRDD) technique assists ophthalmologists by allowing early and precise detection of rare retinal diseases, thus enhancing clinical decision-making and decreasing diagnostic workload. The GAIAFF-RRDD technique integrates advanced pre-processing, augmentation, feature fusion, and classification strategies to improve diagnostic accuracy in retinal image analysis. A conditional generative adversarial network (cGAN) is employed to augment realistic images and overcome the limited availability of images for rare retinal diseases. Furthermore, adaptive feature fusion is performed by combining features extracted from CapsNet, DenseNet-121, and VGG-16 methods. Finally, a variational autoencoder (VAE)-based classification technique is used to classify rare retinal diseases accurately. To ensure improved performance of the GAIAFF-RRDD methodology, an extensive simulation analysis is conducted using the RFMiD dataset. The comparison study of the GAIAFF-RRDD methodology showed an accuracy of 93.92% compared with existing approaches. Therefore, the proposed model is found to be a robust tool for automated and accurate rare disease detection process.