Indywidualna predykcja czasu trwania leczenia w łagodnej padaczce dziecięcej ze złogami centrotemporal: podejście oparte na sieciach morfometrycznej podobieństwa i modelowaniu predykcyjnym

OBJECTIVE: The individual remission timing for children with Benign Childhood Epilepsy with Centrotemporal Spikes (BECTS) is difficult to predict, presenting significant challenges for clinical medication management. This study aimed to investigate alterations in the cortical Morphometric Similarity Network (MSN) in children with BECTS and, on this basis, to construct a Connectome-based Predictive Model (CPM) for the precise prediction of individualized medication duration. METHODS: We employed a dual-center longitudinal design. The study recruited 79 children with BECTS and 72 healthy controls (HC) from Center 1 as the discovery cohort, and 29 children with BECTS from Center 2 as an independent validation cohort. All participants underwent high-resolution T1-weighted imaging. Based on the Desikan-Killiany atlas, multiple morphometric features were extracted from 308 brain regions to construct individual MSNs. We first compared MSN differences between patients with BECTS and HCs. Subsequently, using a leave-one-out cross-validation CPM framework within the discovery cohort, we identified MSN connectivity features associated with medication duration to build a predictive model, which was then tested for generalizability in the independent cohort. RESULTS: Relative to HCs, children with BECTS exhibited significant MSN abnormalities in key brain regions involving the sensorimotor, default mode, and frontoparietal control networks (p < 0.05, Bonferroni-corrected). Leveraging these network anomalies, the CPM successfully extracted predictive features from pre-treatment MSNs, significantly predicting individualized medication duration in the discovery cohort (r = 0.309, p = 0.006), with the positive feature set yielding the best performance (r = 0.325, p = 0.004). The model maintained significant predictive capacity in the independent validation cohort (r = 0.421, p = 0.023). CONCLUSIONS: This study is the first to reveal specific morphometric similarity network abnormalities in children with BECTS and to successfully construct a generalizable connectome-based predictive model based on these findings. The model enables individualized prediction of medication remission time based on pre-treatment brain structural characteristics, providing a potential objective tool for prognostic stratification and precision clinical management of BECTS.