Segmentation of COVID-19 CT Lesions in CT Scans through Transfer Learning

Symeon Psaraftis-Souranis ; Christos Troussas ; Athanasios Voulodimos ; Cleo Sgouropoulou

Computer Science and Information Systems, Tome 22 (2025) no. 1.

Résumé

Since its emergence at the end of 2019, SARS-CoV-2 has infected millions of people worldwide, putting healthcare systems to the test. This has led many researchers to investigate how machine learning can assist clinicians in diagnosing infections caused by SARS-CoV-2. Building upon prior studies, we propose a novel deep learning framework designed for the segmentation of lesions evident in Computed Tomography (CT) scans. For this work, we utilized a dataset consisting of 20 CT scans annotated by experts and conducted training, validation and external evaluation of the deep learning models we implemented, employing a 5-fold cross-validation scheme. When splitting data by slice, our optimal model achieved noteworthy performance metrics, attaining a Dice Similarity Coefficient (DSC) and Intersection over Union (IoU) score of 0.8644 and 0.7612, respectively, during the validation phase. In the external evaluation phase, the model maintained a strong performance with a DSC and IoU score of 0.7211 and 0.5641, respectively. When splitting data by patient, our optimal model achieved a DSC and IoU score of 0.7889 and 0.6540, respectively, during the validation phase. In the external evaluation phase, the model maintained a strong performance with a DSC and IoU score of 0.7079 and 0.5480, respectively. The results, obtained from this research, suggest that the incorporation of transfer learning, combined with appropriate pre-processing techniques, can contribute to achieving state-of-the-art performance in the segmentation of lesions associated with SARS-CoV-2 infections.