Radiomics models with baseline MRI and clinical data to predict target therapy response and high-risk mortality in metastatic GIST

利用基线MRI和临床数据构建放射组学模型,预测转移性GIST患者的靶向治疗反应和高危死亡率。

阅读:2

Abstract

To develop radiomics models based on baseline multi-parametric magnetic resonance imaging (MRI) data and clinical characteristics to predict 6-month progressive disease (PD) status in metastatic gastrointestinal stromal tumor (GIST) patients receiving targeted therapy, enabling long-term prognostication for early stratification of high-risk mortality groups. Eighty-eight metastatic GIST patients undergoing targeted treatment were included in this study and randomly divided into a training cohort and a validation cohort in a ratio of 2:1, comprising 32 disease progression (PD)-positive patients and 56 PD-negative patients. Follow-up computed tomography (CT) or MRI scans obtained 6 months after the baseline MRI were used to determine progressive disease (PD) status according to RECIST 1.1 criteria. Radiomics features were extracted from baseline T2-weighted imaging (T2WI), contrast-enhanced T1-weighted imaging (CE-T1WI), and apparent diffusion coefficient (ADC) sequences. Correlation-based feature selection, information gain, and least absolute shrinkage and selection operator (LASSO) regression were employed in a ten-fold cross-validation to select relevant image features. The radiomics score (Radscore), calculated from the three MRI sequences, along with statistically significant clinical characteristics between the PD-positive and PD-negative groups in both cohorts, were used in multilogistic logistic regression to build the radiomics models. Potential variables for stratifying metastatic GIST patients into distinct mortality risk categories were evaluated using Kaplan-Meier survival analysis, with both dichotomous (Radscore, radiomics predictions and 6-month PD status) and trichotomous (mitotic count and current tumor distribution) classification approaches. The radiomics model, integrating Radscore, mitotic count per 50 high-power fields (HPFs), and current tumor distribution, demonstrated robust discriminatory performance with area under the curves (AUCs) of 0.847-0.974. This integrated model achieved high predictive accuracy for 6-month PD status, yielding classification rates of 94.4% in the training cohort and 84.2% in the testing cohort. The Kaplan-Meier survival analysis demonstrated significant mortality risk stratification (all p < 0.05) for both continuous and categorical variables across cohorts. Specifically, dichotomized variables including Radscore (cutoff > -1.01), radiomics predictions (cutoff > -2.49), and 6-month PD status, along with the trichotomized mitotic count (< 5, 5-10, > 10 per 50 HPFs), effectively discriminated high-risk patients. Univariate Cox regression analysis revealed cohort-specific prognostic patterns, with PD status at 6 months demonstrating the highest predictive accuracy in the training cohort (C-index = 0.782, 95% CI 0.750-0.814), while mitotic count emerged as the strongest predictor in the testing cohort (C-index = 0.819, 95% CI 0.767-0.870). Radiomics models integrating baseline MRI and clinical data provide accurate short-term prognostication of 6-month PD status in metastatic GIST patients, facilitating early risk stratification. However, while the model predicts PD status as a marker of early treatment failure, it does not replace the established prognostic value of PD status itself for long-term survival outcomes. These models may guide personalized therapy adjustments in the short term, but long-term risk assessment should rely on comprehensive clinical-pathological evaluation.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。