Innovative application of artificial intelligence to pre-screen COVID-19 from digital chest radiographs – our experience in a tertiary care setup

Abstract

<b>Aims and Objectives:</b> Using Artificial Intelligence (AI) to triage COVID-19 pattern on digital chest radiographs (CXR) to ensure prompt reporting by imaging experts and ensure initiation of rapid patient management protocols. Materials and Methods: Using Genki AI enabled public health platform (DeepTek Inc) connected to hospital PACS for pre-screening 9098 CXR of 3180 patients [RT-PCR results – ground truth was available for 604 patients (accounting for 1656 radiographs)]. AI Model – Classification and Segmentation Model combined using weighting function. Affected area was quantified by a bounding box (heat map). <b>Results:</b> The AI triage was rapid and happened within a few seconds. The model assessed all the radiographs of the patients and classified each of them into one of two classes: (1) patient likely to have COVID-19 and (2) patient unlikely to have COVID-19. All these images were retrospectively evaluated by an expert radiologist. AI model when compared to RT-PCR as ground truth obtained a sensitivity of 0.67 at a specificity of 0.43. While when AI model was compared to radiologists reading as ground truth, it matched radiologists performance and obtained a sensitivity of 0.87 at a specificity of 0.60. <b>Conclusion:</b> Sophisticated technology like AI can empower healthcare experts to improve work efficiency, minimize errors and reduce burnouts. AI for CXRs through its power of instant pre-screening can reduce burden on the healthcare systems, and facilitate triage in remote areas with no need of a dedicated radiologist in the near future.

Ähnliche Arbeiten