5.AO_884_MAL_Evaluation innovante des atteintes pulmonaires COVID-19 par IA

Abstract

Since the COVID-19 pandemic, rapid diagnosis and monitoring of lung involvement have been essential. The RT-PCR test, although standard, has limitations in terms of delay and accessibility, particularly in sub-Saharan Africa. Thoracic computed tomography (CT) has become a crucial tool for directly visualizing characteristic lung lesions, but its manual interpretation&C suffers from significant variability. Artificial intelligence (AI) applied to imaging overcomes these limitations.

The study conducted at Polyclinique Pasteur in Bamako (229 patients, January 2020–June 2023) compares the quantitative evaluation of lung involvement using the AI software "VCAR Thoracic" to that performed by expert radiologists. This software automatically segments lesions and quantifies anomalies such as ground-glass opacities and consolidations, with 98% sensitivity and 95% specificity.

Results show a similar average lung involvement estimate between AI (35%) and visual analysis (33%) with no statistical difference. The three-level severity classification agrees at 80%, confirming the software's reliability. AI improves evaluation standardization, reduces subjectivity, accelerates analysis, and facilitates patient longitudinal follow-up.

Integration of VCAR Thoracic software optimizes COVID-19 clinical management by providing objective and reproducible quantification, representing a major advance in thoracic radiology for managing emerging pulmonary infections.