Diagnostic accuracy of the automatic quantification software in patients with suspected COVID-19 of the Instituto Nacional de Cardiology Ignacio Chávez

Estefanía Gallego-Díaz, Programa de Imagenología Diagnóstica y Terapéutica, Departamento de Radiología, Instituto Nacional de Cardiología Ignacio Chávez; Grupo CT Scanner, Universidad Nacional Autónoma de México; Ciudad de México, México
César N. Cristancho- Rojas, Departamento de Radiología, Salud Digna, Ciudad de México, México
Sergio A. Criales-Vera, Departamento de Radiología e Imagen, Instituto Nacional de Cardiología Ignacio Chávez; Servicio de Tomografía y PET, Grupo CT Scanner. Ciudad de México, México

Background and Objectives: To establish the diagnostic accuracy of the computed tomography (CT) comparing the probability of COVID-19 pneumopathy, obtained through artificial intelligence (AI) system designed by Siemens Healthineers, and the qualitative evaluation CO-RADS (COVID-19 Reporting and Data System) with the reference standard (RT-PCR), and thus providing the experience of our institution. Methods: An observational, comparative and retrolective study was performed on 280 adult patients with suspected SARS-CoV2 infection, 192 of whom had PCR testing. Diagnostic accuracy information was obtained after comparing the reference standard (RT-PCR) with the CO-RADS performed by observers and the probability of COVID-19 yielded by CT images through AI software. Results: The comparison of COVID-19 probability acquired by AI vs. SARS CoV-2 RT-PCR generated an AUC ROC 0.774 (CI 0.69-0.81) with p = 0.0001. The COVID-19 probability had an acceptable accuracy, with a good PPV 87.80%, but with a poor NPV of 58.80%. The CO-RADS vs. RCP variable got a higher accuracy with much higher sensitivity and specificity values, reaching 91.80% and 88.7% respectively. Conclusion: The comparison between the results obtained by the AI and those referring to the CO-RADS variable showed greater effectiveness in the latter for patients with suspected COVID-19 however, it was possible to document the high impact of the automatic quantification system in the evaluation of these patients since it allows speeding up the radiologist’s assessment and works as a complement in cases of diagnostic doubts.

Keywords: Artificial intelligence. Automatic quantification. COVID-19. CO-RADS. RT-PCR.