Automated Quantitative Ulcer Analysis (AQUA): Diagnosing Organism Types

University Of Michigan At Ann Arbor, Ann Arbor MI

Investigators

Abstract

PROJECT SUMMARY/ABSTRACT Microbial keratitis (MK) is an emergent, painful infection of the cornea and a leading cause of worldwide blindness. A key modifiable driver of vision loss in MK is the lack of a reliable, rapid, and effective strategy to identify the causative organism at disease presentation. Point-of-care tests are not available, are experimental, and are cost-prohibitive. The lack of point-of-care organism identification leads clinicians to be unsure about treatments, affecting patient outcomes. There is a critical need for developing practical, widely available strategies to rapidly identify the causative organism type and to effectively convey the results to eye clinicians. The long-term goal is to create adaptable low-cost strategies that support rapid, equitable MK clinical decision-making. This proposal’s objective is to create: (1) knowledge about MK organism identification, (2) an Automated Quantitative Ulcer Analysis (AQUA) algorithm to classify organism types, and (3) a point-of-care AQUA decision tool to aid clinicians. Our central hypothesis is that a simple platform integrating the AQUA algorithm and decision tool will meet algorithm performance standards and will improve decision making at patient presentation. This hypothesis is supported by preliminary data that has established the AQUA algorithm performance, the targeted performance required to achieve a net benefit for clinicians, and the identification of the domains for an MK decision tool. We will achieve the following Specific Aims: (1) identify relative weights for risk factors associated with organism types, (2) create and validate a novel machine-learning algorithm that identifies MK organism types, (3) establish the drivers of decisions at point-of-care, and (4) build and assess a decision tool that support clinicians. The project will create fundamental knowledge on MK and on decision making for MK, develop and freely-distribute novel machine learning algorithms from a diverse population and develop MK decision tools. This work will create a low cost point-of-care strategy for clinicians who are managing a complex, rapidly-moving, and visually-impactful disease. It will also create open-source algorithms, data, and models for researchers to apply broadly.