Cerebral microstructural alterations as an imaging biomarker for Post-COVID-condition

To develop an imaging biomarker-based approach for the diagnosis of Post-COVID-condition (PCC) at the individual patient level. Magnetic resonance imaging (MRI) data from a prospective cohort of PCC patients (n = 89) were compared with a control group of unimpaired individuals who had contracted coronavirus disease 2019 (COVID-19) in the past (n = 38). Participants were divided into two groups: a training and a test cohort. The macrostructure, diffusion tensor imaging, and multi-shell-based microstructure imaging metrics were extracted using an atlas-based approach. These data were subsequently utilized to train a linear support vector machine (SVM). The efficacy of discrimination between the groups was evaluated for various combinations of input parameters. Upon comparison of the different input combinations, we found the highest area under the receiver operating characteristic curve (AUROC) for microstructural parameters. For the optimal combination of input parameters, an AUROC value of 0.95 with a sensitivity of 94% and a specificity of 85% was achieved, indicating high discriminatory potential but also underscoring the need for further validation given the non-negligible false-positive rate. The atlas regions with the highest discriminatory power include both gray (including multiple cortical areas, putamen and left thalamus) and white matter (including corpus callosum and frontal white matter). The use of a SVM allowed for the differentiation between PCC patients and UPC participants with high sensitivity using microstructural MRI data. While these findings mark a significant step toward a biomarker-based diagnosis of PCC, the moderate specificity and the monocentric design emphasize the need for confirmation in larger and multicentric cohorts before clinical application.