Project Summary: Fifteen percent of people with diabetes suffer from diabetic foot ulcers (DFUs). Within a year of DFU diagnosis, about 17% will need minor amputation and 5% will need a major one. Complications from non-healing infection and ischemia worsen DFU outcomes. Enabling early recognition of infection and ischemia, and providing appropriate therapy, would enhance healing, reduce amputations, and save substantial costs. Effective DFU management requires infection and ischemia status evaluation at multiple time points. Access to wound evaluation and care centers would improve outcomes, especially in high-risk, minority populations. Therefore, there is an urgent need for non-invasive tools for detecting DFU onset of wound infection and ischemia. We organized a new venture, Anxomics whose mission is to bring an improved, home-based diagnostic system for DFUs to market. The company has exclusive access to a deep learning-based image analysis system to accurately segregate wound tissue from normal skin and measure physical parameters, including size, color, and texture. Convolution neural network-based deep-learning models trained on segmented wounds achieved an accuracy of 79.8% in differentiating infected vs. non-infected DFUs based on images in the independent validation set. Similarly, the ischemia phenotype-trained deep learning model achieved 94.81% accuracy. Further testing in an independent pilot study demonstrated accuracies of 88.9% and 94.4% in infection and ischemia identification using images captured by conventional smartphones in standard clinical settings. Dispensing with the need for more complicated and expensive image capture technology facilitates the broad implementation of a remote, home-based, wound management system for DFUs. The platform is being extensively tested using DFU images acquired from people with different skin tones (collected from Emory University Hospital in Atlanta, which serves minorities, and another hospital in north India), taken under different lighting and with different cameras to develop robust prediction models. Our results to date confirm the potential utility of an image-based artificial intelligence technology to offer unprecedented, rapid, and accurate predictive diagnosis of DFU infection and ischemia status. During this Phase 1 STTR project, we propose to optimize and validate our platform for complete analysis of the physical parameters of the wound and infection and ischemia status by completing the following specific aims: Aim 1a: Optimize the parameters for accurate wound segmentation and deep learning (DL) based ischemia and infection prediction on a set of prospectively collected images. (1b) Assess the performance of deep learning-based models in predicting wound infection and ischemia on an independent dataset. Further, we aim to develop a mobile application (DFUCare) for the management of physical, macroscopic, and infection/ischemia data relating to wounds (Aim2).