Rapidly deploying a COVID-19 decision support system in one of the largest Brazilian hospitals

Author(s)

Carmo, Diedre ^[1] ; Campiotti, Israel ^{[1, 2]} ; Rodrigues, Livia ^[1] ; Fantini, Irene ^[1] ; Pinheiro, Gustavo ^[1] ; Moraes, Daniel ^[2] ; Nogueira, Rodrigo ^{[1, 2]} ; Rittner, Leticia ^[1] ; Lotufo, Roberto ^{[1, 2]}

Affiliation

^[1] Univ Estadual Campinas, MICLab, Sch Elect & Comp Engn, Av Albert Einstein 400 Cidade Univ, BR-13083852 Campinas, SP - Brazil

^[2] NeuralMind Inteligencia Artificial, Campinas, SP - Brazil

• Document type: Journal article
• Source: HEALTH INFORMATICS JOURNAL; v. 27, n. 3 JUL 2021.
• Web of Science Citations: 0

Abstract

The COVID-19 pandemic generated research interest in automated models to perform classification and segmentation from medical imaging of COVID-19 patients, However, applications in real-world scenarios are still needed. We describe the development and deployment of COVID-19 decision support and segmentation system. A partnership with a Brazilian radiologist consortium, gave us access to 1000s of labeled computed tomography (CT) and X-ray images from Sao Paulo Hospitals. The system used EfficientNet and EfficientDet networks, state-of-the-art convolutional neural networks for natural images classification and segmentation, in a real-time scalable scenario in communication with a Picture Archiving and Communication System (PACS). Additionally, the system could reject non-related images, using header analysis and classifiers. We achieved CT and X-ray classification accuracies of 0.94 and 0.98, respectively, and Dice coefficient for lung and covid findings segmentations of 0.98 and 0.73, respectively. The median response time was 7 s for X-ray and 4 min for CT. (AU)

• FAPESP's process: 13/07559-3 - BRAINN - The Brazilian Institute of Neuroscience and Neurotechnology
• Grantee: Fernando Cendes
• Support Opportunities: Research Grants - Research, Innovation and Dissemination Centers - RIDC
• FAPESP's process: 19/21964-4 - Lung diseases and COVID diagnosis from computed tomography images using explainable deep learning
• Grantee: Diedre Santos do Carmo
• Support Opportunities: Scholarships in Brazil - Doctorate