A deep-learning model that predicts the spontaneous passage of ureteral stones based on the combination of stone shape and ureter structure
10.3969/j.issn.1002-1671.2024.10.020
- VernacularTitle:利用结石形态与输尿管结构联合预测输尿管结石自发排出的深度学习模型
- Author:
Cong WANG
1
;
Yumeng ZHANG
;
Chao ZHANG
Author Information
1. 山东大学第二医院泌尿外科,山东 济南 250033
- Keywords:
ureteral stone;
artificial neural network;
deep-learning;
artificial intelligence
- From:
Journal of Practical Radiology
2024;40(10):1667-1670
- CountryChina
- Language:Chinese
-
Abstract:
Objective To investigate the prediction of spontaneous passage of ureteral stones by deep-learning models using the combination of stone shape and ureter three-dimensional structure.Methods A total of 280 patients who received stone-passing therapy were analyzed retrospectively.The three-dimensional shape of the ureteral stone and the three-dimensional structure of the ipsilateral ureter were marked on CT scan.All patients were randomly divided into a training set(200 cases)and a test set(80 cases).Two deep-learning models based on convolutional neural network(CNN)was developed,one(CNN-S)only used stone shape as input,and the other(CNN-SU)used both stone shape and ureter three-dimensional structure as input.The training set and validation set were used to train and validate these two models,and the test set was used to exam the precision of these two models.In addition,the maximum stone diameter＜5 mm was used as the expulsion standard for comparison.Results Stone diameter classification had an accuracy of 71.3％,sensitivity of 46.7％,and specificity of 86.7％.Single input model used stone shape had an accuracy of 85.0％,sensitivity of 86.7％,and specificity of 84.0％.Double input model had an accuracy of 90.0％,sensitivity of 93.3％,and specificity of 88.0％.Both models achieved higher accuracy than diameter classification,the difference was statistically significant.Double input model had higher accuracy,sensitivity and specificity than single input model,but the difference was not significant.Conclusion Deep-learning models can predict the spontaneous passage of ureteral stone more accurately than diameter classification with stone shape input alone.Combining stone shape and ureter three-dimensional structure can further improve accuracy.These models can be very helpful for clinical decision-making of ureteral stone treatment.
