RayEye

Hey there, Amritesh here 👋 We're excited to launch a product we've been working on for some time. Why is it needed? Most COVID-19 patients are diagnosed with pneumonia and showcase some characteristic CT imagine patterns. Even though, with current models, chest CT scans (or XRays) could be non-specific for COVID-19 detection, it can be used as a major evidence of clinical diagnosis for 2 reasons- 1. First, RT-PCR test results can be affected by sampling errors. There have also been cases where RT-PCR is negative and CT scans show signs of COVID and RT-PCR later turns positive. 2. Secondly, RT-PCR tests are scarce and the situation may get worse as we get more cases. In such situation, our model for CT scans can detect COVID-19 related pneumonia easily and quickly, which even though might not be considered a diagnosis at this stage, but can help fast-track quarantine, contact tracing and prioritising RT-PCR test itself for patients with higher possibility of COVID-19. How it works? Our AI model differentiate healthy CT scan from COVID-19 based on consolidation pattern of ground glass opacity(GGO). Although the consolidation is similar for Influenza, our deep learning model is able to identify the subtle differences and also taking into account other factors such as number extent and density of GGO. System Performance We currently have achieved >95% accuracy for CT scan conditions like GGO based on 929 CT scans. We need to collect more data to be able to accurately and reliably diagnose COVID-19 using CT scans. For COVID diagnosis using XRays, accuracy = 91%, sensitivity = 97% on dataset of ~13000 Xrays out of which 183 were COVID-19 positive. We are also able to differentiate between COVID-19 and other types of Pneumonia. At this point, we are accurately predicting conditions like GGO, consolidation and pleural effusion which can be used to classify risk level for COVID-19 and also for tracking recovery of COVID patients. In future, we will also build COVID-19 diagnosis capability based on symptoms and medical history. Me, Justin, Ronald and Maheshwar will be here replying to your comments and questions. Please try it out and let us know how it goes. For further details, please refer to research section on our website. The Rayeye team P.S. This is an open source project. Please reach out to us if you want to contribute or interested in using.