Development of a multi-parameter MRI-based habitat radiomics model for differential diagnosis of vestibular schwannoma and meningioma in the cerebellopontine angle: a multi-center study.
Authors
Affiliations (7)
Affiliations (7)
- Department of Neurosurgery, Guangyuan Central Hospital, Guangyuan, Sichuan, 628000, China.
- Department of Neurosurgery, Dazhou First People's Hospital (Sichuan Provincial People's Hospital Chuandong Hospital), Dazhou, Sichuan, 635000, China.
- Department of Radiology, Xi'an Qinhuang Hospital, Xi'an, Shanxi, 710600, China.
- Department of Pain Medicine, Guang'an People's Hospital, Guang'an, Sichuan, 638000, China.
- Department of Neurosurgery, Affiliated Hospital of North Sichuan Medical College, Nanchong, Sichuan, 637000, China.
- Department of Neurosurgery, Guangyuan Central Hospital, Guangyuan, Sichuan, 628000, China. [email protected].
- Department of Neurosurgery, Beijing Anzhen Nanchong Hospital of Capital Medical University & Nanchong Central Hospital, Nanchong, Sichuan, 637000, China. [email protected].
Abstract
Vestibular schwannomas (VS) and meningiomas (MM) are the most common benign tumors in the cerebellopontine angle (CPA), often presenting with similar clinical symptoms. Accurate preoperative differentiation is essential for appropriate treatment planning. To develop an MRI-based habitat radiomics model for the differential diagnosis of VS and MM in the CPA, integrating multi-parameter MRI data and clinical features. This retrospective study included 176 patients diagnosed with CPA VS or MM from three medical centers: Guangyuan Central Hospital, Nanchong Central Hospital, and Affiliated Hospital of North Sichuan Medical College. 101 patients (68 VS, 33 MM) met the inclusion criteria. Preoperative MRI scans (T2WI and CE-T1WI) were analyzed using a sliding window method for feature extraction, followed by image preprocessing and Gaussian Mixture Model (GMM) clustering for habitat subregion segmentation.Habitat radiomics models based on T2WI, CE-T1WI, and fused T2WI & CE-T1WI were constructed respectively, and a combined habitat-clinical model was developed by integrating clinical features, and the models were compared across multiple different machine learning classifiers. Multi-parameter models, combining T2WI and CE-T1WI features, outperformed single-parameter models, achieving AUC values of 0.987 in the training set and 0.974 in the test set. A combined model integrating habitat radiomics and clinical features showed superior diagnostic accuracy. Decision Curve Analysis and calibration curves confirmed its clinical utility and reliability. MRI-based habitat radiomics, when integrated with clinical features, provides a robust, noninvasive method for differentiating CPA VS from MM, offering potential for personalized treatment strategies.