A multicenter study on diagnosing clinically significant prostate cancer using a deep learning classification model based on biparametric MRI
10.3969/j.issn.1002-1671.2025.07.018
- VernacularTitle:基于双参数MRI的深度学习分类模型诊断临床显著性前列腺癌的多中心研究
- Author:
Lin LI
1
;
Man LI
;
Saiqun LÜ
;
Jieke LIU
;
Shengbin DENG
;
Qiang ZHANG
;
Tao PENG
Author Information
1. 成都大学附属医院放射科,四川 成都 610081
- Publication Type:Journal Article
- Keywords:
clinically significant prostate cancer;
biparametric magnetic resonance imaging;
deep learning classification model;
multicenter study
- From:
Journal of Practical Radiology
2025;41(7):1163-1167
- CountryChina
- Language:Chinese
-
Abstract:
Objective To investigate the classification capability of a deep learning classification model based on biparametric mag-netic resonance imaging(bpMRI)for clinically significant prostate cancer(csPCa)and clinically insignificant prostate cancer(cisPCa).Methods A retrospective analysis was conducted on the data of 565 prostate bpMRI patients.A deep learning classification model was established for csPCa.The patients were randomly divided into training set(452 cases)and internal test set(113 cases)at a ratio of 8︰2.Internal validation was performed,followed by external validation(external validation set)using data from 120 patients across four different hospitals.The area under the curve(AUC)of the receiver operating characteristic(ROC)curve,F1 score,precision,sensi-tivity,specificity,accuracy,and calibration curves were used to evaluate the model.Decision curve analysis(DCA)was also applied to assess the clinical benefit of the model.Results The deep learn-ing classification model for csPCa classification demonstrated the following performance across the training set,internaltest set,and external validation set:sensitivity of 0.986,0.887,and 0.750;specificity of 0.967,0.850,and 0.976;precision of 0.963,0.839,and 0.818;accuracy of 0.974,0.862,and 0.792;F1 score of 0.974,0.862,and 0.783;and AUC of 0.998,0.896,and 0.883,respec-tively.The calibration curves for all three datasets showed high consistency between predicted and actual probabilities.DCA indicated that the highest net benefit threshold probabilities for the training set,internal test set,and external validation set were 0.2-0.7,0.2-0.6,and 0.2-0.5,respectively.Conclusion The deep learning classification model demonstrated excellent performance in classifying csPCa and exhibited good generalizability,which is worhty of clinical application.
