MRI-based multiregional radiomics for preoperative prediction of Ki-67 expression in meningiomas: A two-center study

Ming Luo^1,2Guihan Lin^1,2Duoning Chen^1,2Weiyue Chen^1,2Shuiwei Xia^1,2Junguo Hui^1,2Pengjun Chen^1,2Minjiang Chen^1,2Wangyang Ye^1,2Jiansong Ji^1,2*

• ¹Lishui Central Hospital, Lishui, China
• ²Fifth Affiliated Hospital of Wenzhou Medical University, Lishui, Zhejiang Province, China

Background: High expression of Ki-67 in meningioma is significantly associated with higher histological grade and worse prognosis. The non-invasive and dynamic assessment of Ki-67 expression levels in meningiomas is of significant clinical importance and is urgently required. This study aimed to develop a predictive model for the Ki-67 index in meningioma based on preoperative magnetic resonance imaging (MRI).Methods: This study included 196 patients from one center (internal cohort) and 92 patients from another center (external validation cohort). Meningioma had to have been pathologically confirmed for inclusion. The Ki-67 index was classified as high (Ki-67 ≥5%) and low (Ki-67 <5%). The internal cohort was randomly assigned to training and validation sets at a 7:3 ratio. Radiomics features were selected from contrast-enhanced T1-weighted MRI using the least-absolute shrinkage and selection operator and random forest methods. Then, we constructed a predictive model based on the identified semantic and radiomics features, aiming to distinguish high and low Ki-67 expression. The model's performance was evaluated through internal cross-validation and validated in the external cohort. Results: Among the clinical features, peritumoral edema (p = 0.001) and heterogeneous enhancement (p = 0.001) were independent predictors of the Ki-67 index in meningiomas. The radiomics model using a combined 8 mm volume of interest demonstrated optimal performance in the training (area under the receiver operating characteristic curve [AUC] = 0.883) and validation (AUC = 0.811) sets. A nomogram integrating clinical and radiomic features was constructed, achieving an AUC of 0.904 and enhancing the model's predictive accuracy for high Ki-67 expression.Conclusions: This study developed clinical-radiomic models to non-invasively predict Ki-67 expression in meningioma and provided a novel preoperative strategy for assessing tumor proliferation.