Generative AI Model Development for Enhancement of Real-time Medical Image-based Training Datasets

The evolution of artificial intelligence (AI) models is driven and shaped by training data. In industries where data is naturally sparse, AI systems face a significant challenge due to their intensive data needs. Anxieties about underprivileged groups' present and future healthcare AI bias stem from the difficulty of curating training data in the healthcare industry. To reduce reliance on huge data, built an autoencoder to augment and expand intentionally sparse datasets, as detailed in this paper. A computational investigation using publicly available data. Patients' ages ranging from 45 to 75 from Africa, Europe, India, and Bangladesh were included in six open-source datasets. The presented architecture uses actual patient imaging data to create synthetic pictures. For this case study, they utilised an autoencoder to augment existing public datasets of optic disc images for training purposes and assessed how well these datasets could train AI models to identify glaucomatous optic neuropathy. An evaluation of the glaucoma detector's performance was conducted using the area under the receiver operating characteristic curve (AUC). Better detection performance is indicated by a higher AUC. The results demonstrate that by supplementing datasets with autoencoder-generated synthetic pictures, better training sets were produced, ultimately leading to enhanced AI model performance. Beyond health care, this study may boost AI adoption in all data-challenged domains, which is a major concern due to the ever-increasing database volume as well as necessary for AI building.