Developing an interpretable clinical-radiomics machine learning model using whole transition zone MRI analysis for improving diagnosis of transition zone prostate cancer

利用全移行区MRI分析,开发可解释的临床放射组学机器学习模型,以提高移行区前列腺癌的诊断率

阅读:2

Abstract

OBJECTIVES: Transition zone prostate cancer (TZ-PCa) presents significant diagnostic challenges due to overlapping imaging features with benign prostatic hyperplasia (BPH). This study aimed to develop and externally validate an interpretable clinical-radiomics model that integrates biparametric MRI (bp-MRI; T2-weighted imaging (T2WI) and apparent diffusion coefficient (ADC)) features with clinical variables to improve the diagnostic accuracy of TZ-PCa. METHODS: A total of 280 pathologically confirmed cases from two institutions were retrospectively analyzed. Patients from Center A (n=203) were divided into a training set (n=142) and an internal validation set (n=61), while patients from Center B (n=77) constituted an external validation set. The whole transitional zone on the slice corresponding to the tumor's largest diameter was delineated as a single-slice region-of-interest (ROI). Radiomics features were extracted and used to train six machine learning algorithms to construct single-sequence (T2WI or ADC) and combined-sequence (ADC+T2WI) models. The best radiomics model was then combined with independent clinical characteristics to construct a clinical-radiomics model. Model performance was evaluated by Receiver Operating Characteristic (ROC) analysis, and clinical utility was assessed with calibration and decision curve analyses (DCA). The interpretability of the optimal model was further examined using Shapley Additive Explanation (SHAP). RESULTS: Multivariate logistic regression analysis identified PI-RADS score (odds ratio (OR)=3.47, 95%CI 1.90~6.35, P<0.001) and total prostate specific antigen (tPSA) (OR = 1.06, 95%CI 1.01~1.12, P=0.020) as independent clinical predictors. The support vector machine (SVM) radiomics model using combined ADC+T2WI features achieved AUCs of 0.865 (training) and 0.850 (internal validation). The clinical-radiomics model yielded AUCs of 0.963, 0.889, and 0.829 in the training, internal validation, and external validation sets, respectively. SHAP analysis identified T2-wavelet-LLH_glszm_SmallAreaLowGrayLevelEmphasis as the most crucial feature. CONCLUSION: The proposed clinical-radiomics model demonstrated the best diagnostic performance for differentiating TZ-PCa from BPH across bio-centers. Combining the SHAP algorithm with the model enhances interpretability and may assist clinicians in making more precise diagnostic and treatment decisions.

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。