Prediction of biochemical recurrence after radical prostatectomy based on MRI intratumoral and peritumoral radiomics combined with pathological parameters
10.3969/j.issn.1002-1671.2025.11.015
- VernacularTitle:基于MRI瘤内瘤周影像组学联合病理参数预测根治性前列腺切除术后生化复发
- Author:
Tong LUO
1
;
Ling ZHAO
1
;
Weibin CHEN
1
Author Information
1. 华北理工大学附属医院医学影像中心,河北 唐山 063000
- Publication Type:Journal Article
- Keywords:
prostate cancer;
biochemical recurrence;
magnetic resonance imaging;
radiomics
- From:
Journal of Practical Radiology
2025;41(11):1830-1834
- CountryChina
- Language:Chinese
-
Abstract:
Objective To investigate the efficacy of intratumoral and peritumoral radiomics features besed on biparametric MRI combined with pathological parameters in predicting biochemical recurrence(BCR)after radical prostatectomy(RP).Methods A total of 295 prostate cancer(PCa)patients who underwent RP were retrospectively selected and divided into training set(n=206)and test set(n=89)in a 7∶3 ratio.Subgroups were stratified based on prostate specific antigen(PSA)results during 24-month postopera-tive follow-up.Radiomics features were extracted and integrated from both intratumoral and peritumoral 3 mm dual-regions on T2WI and apparent diffusion coefficient(ADC)sequences.Feature selection was performed using independent sample t-tests,maximum relevance and minimum redundancy(mRMR)algorithm,and least absolute shrinkage and selection operator(LASSO)regression.Binary logis-tic regression analysis was used to screen the independent pathological risk factors.Radiomics models(intratumoral,peritumoral,and combined)and a pathological model were constructed based on the Light gradient boosting machine(LightGBM)algorithm.A nomo-gram model was developed by integrating the pathological model with the combined model.The model efficacy was evaluated via the receiver operating characteristic(ROC)curves,and the DeLong's test and net reclassification improvement(NRI)were used to com-pare the efficacy differences among the models.Results Finally,the optimal feature numbers for the intratumoral,peritumoral,and combined models were 5,6,and 6,respectively.In the test set,the combined model demonstrated significantly higher area under the curve(AUC)than the intratumoral model(0.905 vs 0.815,P＜0.05).Ki-67,extracapsular extension,and positive surgical margin were identified as independent pathological predictors of BCR(P＜0.05).The nomogram model demonstrated significantly higher AUC than the pathological model(0.931 vs 0.786,P＜0.05),with no significant difference versus the combined model(0.931 vs 0.905,P＞0.05).Conclusion The nomogram model integrating intratumoral,peritumoral radiomics features with pathological parameters can effectively predict the occurrence of BCR within 2 years after PCa surgery.
