Multiparametric-MRI Habitat Radiomic Analysis for Discriminating Pathological Types of Brain Metastases

Jinling Zhu¹Xin Xie¹Jixuan Deng¹Ruizhe Xu²Li Zou²Ye Tian²Wu Cai¹Bo Zhang^1*

• ¹Department of Radiology, Second Affiliated Hospital of Soochow University, Suzhou, China
• ²Department of Radiotherapy & Oncology, Second Affiliated Hospital of Soochow University, Suzhou, China

Background: Early identification of the primary tumor type in brain metastases (BMs) is crucial for developing effective treatment strategies. This study aimed to evaluate the potential of multiparametric MRI (mpMRI)-based habitat radiomic analysis in differentiating the histopathological types of BMs. Materials and methods: Pre-treatment MR images from 328 BMs patients at a single center were retrospectively collected and randomly divided into a training set (229 cases) and a test set (99 cases). Tumor regions were manually segmented on contrast-enhanced T1-weighted images (CE-T1WI), and the K-means clustering algorithm was employed to classify the tumor into four distinct sub-regions. Radiomics features were extracted separately from each sub-region to construct the habitat model. For comparison, whole-tumor-based radiomic features were also extracted. Four predictive models were developed and compared: the habitat model, the traditional radiomics model, the clinical model, and the combined model (integrating habitat radiomics features with clinical variables). Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), as well as accuracy, sensitivity, and specificity. Results: The habitat model exhibited robust predictive performance in both the training and test sets (AUC 0.965/0.888, accuracy 0.876/0.835). The traditional radiomics model showed slightly lower performance (AUC 0.984/0.884, accuracy 0.866/0.754). The clinical model demonstrated the lowest diagnostic accuracy (AUC 0.788/0.716, accuracy 0.731/0.653). The combined model achieved the highest overall performance on the test set, with an AUC of 0.939, an accuracy of 0.845, and a micro-average F1-score of 0.796. Class-specific analysis revealed the following F1-3 scores: 0.200 for gastrointestinal cancer, 0.333 for breast cancer, and 0.874 for lung cancer metastases. Conclusions: This study establishes that while habitat radiomics shows potential for classifying brain metastases, its current performance is constrained by class imbalance and scanner heterogeneity. Consequently, our primary contribution lies in providing a critical baseline and a clear roadmap, prioritizing data-centric solutions as the essential next step for the field.