Przewidywanie wyników leczenia chirurgicznego u pacjentów z lekooporną epilepsją skroniową za pomocą analizy obrazów radiomicznych hipokampa z hybrydowego badania PET/MRI

PubMed➕ 20.11.2026BMC Med Imaging

Predicting post-surgical seizure outcomes in drug-resistant temporal lobe epilepsy using hippocampal radiomics from F-FDG hybrid PET/MRI

W skrócie

Badacze opracowali nową metodę, która łączy wyniki kilku nowoczesnych badań obrazowych (PET i MRI) wraz z danymi klinicznymi pacjenta, aby dokładniej przewidzieć, czy operacja pomoże w kontrolowaniu napadów epilepsji u osób, u których leki nie działają. Metoda ta okazała się bardziej dokładna niż tradycyjne podejście oparte tylko na danych klinicznych. Wyniki sugerują, że analiza obrazów może pomóc lekarzom lepiej wybrać pacjentów do operacji i poprawić ich rokowanie przed zabiegiem.

Oryginalny abstract (angielski)

OBJECTIVE: To develop multivariate hippocampal radiomics models using multimodal F-FDG hybrid PET/MRI and clinical variables to predict post-surgical seizure outcomes in patients with drug-resistant temporal lobe epilepsy (TLE). METHODS: We retrospectively analyzed 117 drug-resistant TLE patients from Xuanwu Hospital (August 2015-March 2023) who underwent preoperative F-FDG hybrid PET/MRI, surgical intervention, and one-year postoperative follow-up using Engel classification. Structural (3D-T1 and T2-FLAIR) and metabolic (F-FDG PET) images were obtained. Hippocampal subfields were automatically segmented using FreeSurfer on 3D T1 images, with F-FDG PET and T2-FLAIR images registered accordingly. Radiomics features were extracted from the hippocampal subfields. Feature selection was performed using the minimum redundancy maximum relevance algorithm and the least absolute shrinkage and selection operator regression methods to calculate radscores. Radiomics-clinical and clinical models was constructed and model performance was evaluated using receiver operating characteristic analysis and decision curve analysis. Patients were divided into hippocampal sclerosis (HS) and non-HS groups based on histology and mediation analysis explored relationships among radscore, HS, and Engel classification. RESULTS: The radiomics model combining all three imaging sequences outperformed models using single or dual modalities. In the training set, the combined F-FDG PET/MRI radiomics-clinical model demonstrated higher discrimination than the clinical model alone (AUC = 0.915 vs. 0.716, p = 0.004). In the test set, the combined model showed a numerically higher AUC (0.819 vs. 0.743); however, this difference did not reach statistical significance (p = 0.661). Decision curve analysis shown that the combined F-FDG PET/MRI radiomics-clinical model provides higher clinical benefits in all risk threshold compared with the clinical model. The radscore was significantly higher in the HS group than in the non-HS group (t = 2.308, p = 0.023). Mediation analysis shows that radscore is a significant mediator between HS and Engel classification (proportion of indirect effect = 45.8%, p = 0.028). CONCLUSION: The combined F-FDG PET/MRI radiomics-clinical model enhances the prediction of post-surgical seizure outcomes in drug-resistant TLE patients. Radiomics features may reflect the pathological state of HS, aiding in preoperative prognostication and clinical decision-making.

Metadane publikacji

Journal
BMC Med Imaging
Data publikacji
26.08.2026
PMID
43058531
DOI
10.1186/s12880-026-02708-2
Autorzy
Hu J, Yao C, Wang J, Zhang S, Cui B, Wang Z, Hou Y, Lu J
Słowa kluczowe
18F-FDG PET/MR, Engel classification, Hippocampus, Radiomics, Temporal lobe epilepsy
Źródło
PubMed