Arterial spin labeling as a promising alternative to FDG-PET for clinical diagnosis of patients with disorders of consciousness

Junrong Han,Niall W. Duncan,Likai Huang,Chia Lee,Che-Ming Yang,Yen-Chien Wu,David Yen-Ting Chen,Timothy J. Lane

DOI: https://doi.org/10.1101/2024.08.07.24311648

2024-08-08

Abstract:Objective: To evaluate the potential of arterial spin labeling (ASL) as an alternative to FDG-PET in the diagnosis of disorders of consciousness (DOC), we conducted a comparative study of the two modalities. Methods: A total of 36 DOC patients (11 female; mean age = 49.67 ± 14.54 years) and 17 healthy control (HC) participants (9 female; mean age = 31.9 ± 9.6 years) underwent both FDG-PET scans that measure metabolism via glucose uptake and ASL scans that measure cerebral blood flow (CBF). CBF and metabolism in DOC and HC were compared, globally and for seven functional networks. Comparability of the two modalities was estimated using Spearman partial correlation for both global and network levels. Furthermore, a support vector machine (SVM) algorithm was used to train two classifiers to distinguish DOC patients from HC with normalized CBF or metabolism values from the seven networks. Performance of these classifiers was first evaluated through leave-one-subject-out (LOSO) cross-validation within their respective modalities. Subsequently, cross-modal validation was conducted: testing the ASL-trained classifier with PET data (ASL-to-PET validation) and vice versa (PET-to-ASL validation). Performance of each classifier was assessed using receiver operating characteristic (ROC) analysis, with area under the curve (AUC) as the metric. Results: Both modalities showed agreement in decreased CBF and metabolism in DOC patients compared to HC, at both global and network levels. The global brain and most networks showed significant positive partial correlation between CBF and metabolism. SVM classifiers, utilizing activity from seven networks as features, performed well in the both LOSO cross-validation (ASL-trained classifier: accuracy = 83.02%, AUC = 0.95; PET-trained classifier: accuracy = 98.12%, AUC = 0.98) and cross-modal validation (ASL-to-PET validation: accuracy = 84.90%, AUC =0.92; PET-to-ASL validation: accuracy = 73.58%, AUC = 0.93). Conclusion: Our results demonstrate that ASL provides information comparable to FDG-PET for DOC patients; moreover, classifiers trained on ASL data perform comparably well to those trained on FDG-PET data. Therefore, ASL could be a valuable alternative to FDG-PET in the clinical diagnosis of DOC, especially in light of its advantages: ease of acquisition, avoidance of radiation exposure, brevity of scanning time, and lower-cost.

Radiology and Imaging