基于扩散加权MRI的栖息地影像组学预测晚期高级别浆液性卵巢癌PARP抑制剂耐药

RATIONALE AND OBJECTIVES: To develop a habitat radiomics model based on diffusion-weighted magnetic resonance imaging (DW-MRI) for predicting the PARPi resistance in patients with advanced high-grade serous ovarian cancer (HGSOC).

MATERIALS AND METHODS: In this retrospective study, a total of 157 patients with histopathologically confirmed HGSOC from two centers were included and divided into training (n = 111) and testing (n = 46) cohorts. Gaussian mixture model algorithm was used to identify different habitats and then 851 features were extracted from each sub-region. Spearman correlation analysis, univariate Cox regression, and Lasso-Cox regression algorithms were used to select the optimal feature set. Subsequently, the clinical, habitat, and combined models were constructed respectively to predict patients' progression-free survival (PFS). The model performance was evaluated using C-index, the area under the time-dependent receiver operating characteristic (ROC) curve, calibration, and clinical utility.

RESULTS: Habitat radscore was an independent predictor of PARPi resistance in advanced HGSOC. By integrating habitat radscore with the clinical factor, the combined model showed the best performance. The model achieved favorable AUC values of 0.763, and 0.749 for predicting disease progression at 18, and 24 months, respectively, which was significantly higher than those of the clinical model. Patients stratified using the combined model showed significant difference in PFS between the low- and high-risk groups (P < .05 in both cohorts).

CONCLUSION: The combined model integrating habitat radiomics features and clinical factor may help predict PARPi resistance in advanced HGSOC receiving first-line PARPi maintenance treatment.