Executive Summary : | Glaucoma is the second leading cause of blindness in the world. According to a report, in 2010, the number of glaucoma patients in India was as high as 11.2 million. A population-based study in south India also revealed that nearly 10.3% of the subjects had angle-closure glaucoma. Fortunately, pharmacological and surgical solutions currently exist to limit glaucoma progression if detected at an early stage. However, providing an accurate glaucoma diagnosis is a complex endeavor that is time-consuming, heavily dependent on a clinician’s experience, and requires multiple clinical tests. In a developing country like India, where the prevalence of glaucoma is 1.25-2.56%, these visits and diagnostic tests become a substantial socioeconomic burden. It forces many elderly patients to go untreated till they reach the severe stage of glaucoma or start to lose their peripheral vision. The power of artificial intelligence (AI) can be leveraged here to examine the glaucomatous biomarkers autonomously and to assist clinicians in early-stage diagnosis and prognosis. This research aims to design AI-powered tools to extract clinically validated structural biomarkers of the optic nerve head (ONH) from optical coherence tomography (OCT) images and develop standalone software to assist clinicians in glaucoma diagnosis. Training an AI network to identify structural biomarkers is a highly complex task that requires extensive knowledge of structural phenotypes of the glaucomatous ONH and a deep understanding of network architecture and design. At first, a deep learning network with multiple convolution layers will be designed for segmenting neural and connective tissue layers of the ONH. OCT images often contain a lot of noise, blood vessel shadows, and image artifacts. Therefore, advanced denoising and segmentation networks, such as GAN, Trans-Unet, and Unet++, will be customized and tested on these images. Once the segmentation network is ready, an algorithm will extract biomarkers of the ONH from the segmented images that will be treated as an input to a second AI network for diagnosis. The results will be compared with the actual diagnosis by clinicians. Standalone software with a user interface will be developed where clinicians can upload images obtained directly from commercial OCT machines and can visualize the biomarkers and glaucomatous probability. The software will assist clinicians in visualizing the 3D OCT images from different prospective and examining them. The proposed project will have great translational potential with implementable public health solutions using AI upon successful completion. This project also focuses on the “Aatma Nirbhar Bharat Abhiyan” footprint. Starting from data collection, data analytics, modeling of AI tools, and preparing a deliverable standalone tool will comprise a stepping stone to develop a self-reliant diagnostic tool. |