Applications of artificial intelligence to the diagnostic evaluation of infectious keratitis

PROJECT SUMMARY/ABSTRACT This K23 proposal aims to develop and evaluate applications of artificial intelligence (AI) to the diagnostic investigation of infectious keratitis, a major cause of blindness worldwide. This will be accomplished through three specific aims: 1) Develop and evaluate an AI model to identify the etiology of culture-proven infectious

keratitis from an existing database of clinical photographs; 2) Externally validate model performance in a real- world, population-based sample of corneal ulcers; and 3) Develop and evaluate an additional AI model for automated microscopic diagnosis of fungal keratitis. The AI model developed in SA#1 will be trained using a

clinical photography database (the Culture Positive Ulcer Database) collated from several NIH funded clinical trials for infectious keratitis (SCUT, MUTT I & II, CLAIR, and MALIN) conducted over the past several decades as part of the international collaboration between the Francis I. Proctor Foundation and Aravind Eye Hospital in

India. This model's performance will be compared against human experts on culture-proven cases of infectious keratitis. A second repository of imaging and clinical data from corneal ulcers (the MADURAI database) currently in development will be used to externally validate the AI model developed in SA#1 (by estimating its

sensitivity and specificity in a real-world sample) and to train the AI model in SA#3. To accomplish these research goals, we have established an international collaboration between the Casey Eye Institute, the Proctor Foundation, and Aravind. This provides an unprecedented opportunity to leverage the expertise of my

mentors at Casey in artificial intelligence and computer vision-enabled diagnosis of ophthalmic diseases, the expertise of the world-class faculty at Proctor in epidemiology, biostatistics, and infectious keratitis, and the unparalleled volume of infectious keratitis and infrastructure for data collection at Aravind. This collaboration

will facilitate the development of carefully designed and validated AI models which will guide earlier directed antimicrobial therapy and improve visual outcomes in infectious keratitis. My primary career goals are to establish myself as an independent clinician scientist performing research at the interface of technological innovation and international public health. My MPH, medical training, and

research experience have allowed me to develop a strong foundation in public health, the clinical and surgical management of corneal infections, and medical informatics. Over the past nine months of K12 support I have begun developing expertise in machine learning and data science, establishing a foundation which I will build

upon during this K23 award period. The successful application of AI to health care problems requires a multidisciplinary approach involving clinicians, AI methodologists, informaticists, and public health experts. This K23 will allow me to build skills and expertise in each of these disciplines and become well positioned to lead

this movement in the coming years.