Cerebral blood flow alterations and machine learning classification in chronic low back pain using multi-PLD ASL radiomics.

Chronic low back pain (CLBP) is a prevalent condition with unclear pathophysiology and substantial socioeconomic burden. Cerebral blood flow (CBF) alterations have been implicated in CLBP, yet previous arterial spin labeling (ASL) studies using single post-labeling delay (PLD) have yielded inconsistent results. In this study, multi-PLD ASL was combined with machine learning to characterize CBF alterations in CLBP and to explore their classification feasibility. Seventy-eight patients with CLBP and seventy-eight age- and sex-matched healthy controls underwent multi-PLD ASL scanning. Voxel-wise comparisons of normalized CBF were performed, followed by correlation analyses with clinical measures. Radiomics features extracted from brain regions showing significant CBF differences were used to construct machine learning classification models via a rigorous nested cross-validation and LASSO feature selection framework. Compared with healthy controls, patients with CLBP exhibited significant hyperperfusion in the right lingual gyrus and right thalamus. CBF values in the right lingual gyrus were positively correlated with Oswestry Disability Index scores, while thalamic CBF was positively correlated with pain intensity. Among the evaluated models, the XGBoost classifier achieved the best performance, with an area under the curve of 0.842 (95% CI: 0.774-0.901). These findings indicate that region-specific CBF alterations are closely associated with pain severity and functional impairment in CLBP. Machine learning analysis of CBF radiomic features shows potential discriminative performance in identifying patients with CLBP.