Multimodal Artificial Intelligence to Predict Glaucomatous Progression and Surgical Intervention

点击翻译按钮获取中文摘要

Project Summary The overall objective of this proposal, “Multimodal Artificial Intelligence to Predict Glaucomatous Progression and Surgical Intervention”, is to use multimodal artificial intelligence (AI) and deep learning strategies to predict which glaucoma patients will need glaucoma surgery and which are likely to have progressive visual field loss in the future. This study is designed to leverage longstanding well characterized clinical and research cohorts of glaucoma patients and its validated decision support AI infrastructure to predict which glaucoma patients will progress and which will need surgery. The proposal includes the following two Specific Aims. Aim 1 will use baseline electronic health records (EHR), optic nerve head (ONH) optical coherence tomography (OCT) imaging, visual field (VF) data, intraocular pressure (IOP) and central corneal thickness (CCT) in a multimodal DL model to predict the likelihood of surgical intervention for glaucoma. Aim 2 will use baseline EHR, ONH OCT imaging, VF data, IOP and CCT in a multimodal DL model to predict the likelihood of fast glaucomatous visual field progression. To address these aims, existing data from glaucoma patients 1) enrolled in the National Eye Institute funded Diagnostic Innovations in Glaucoma Study (DIGS 1995-present) and African Descent and Glaucoma Evaluation Study (ADAGES 2009-2021), and 2) managed at the UCSD Viterbi Family Department of Ophthalmology will be used in the AI model development and testing. We will also leverage UCSD's existing cloud-based AI pipeline to build a glaucoma-specific platform to train, test and in the future, update the deep learning models developed. In the future, this infrastructure can be used to support randomized clinical trial testing of AI guided glaucoma management and enable real-time decision support for clinicians.