Shen Dan, Yang Tianxiong, Ma Tao, Yang Wenzeng, Li Hongmei, Cui Zhenyu
Dan Shen Department of Urology, Affiliated Hospital of Hebei University, Baoding 071000, Hebei, China.
Tianxiong Yang College of Clinical Medicine, Affiliated Hospital of Hebei University, Baoding 071000, Hebei, China.
Pak J Med Sci. 2025 Jul;41(7):1918-1924. doi: 10.12669/pjms.41.7.11360.
To explore the capability and clinical application potential of the Faster Region-based Convolutional Neural Network (Faster R-CNN), an Artificial intelligence algorithm, in identifying the composition of urinary calculi from CT images.
This was a retrospective study. Data from 776 patients with urinary calculi treated at the Affiliated Hospital of Hebei University from August 2020 to December 2023 were collected. Patients with simple calculi were randomly divided into a model construction group and validation Group-I at a 5:1 ratio, while 60 cases of mixed calculi were randomly selected to form validation Group-II. The model construction group was employed to construct and test the performance of the Faster R-CNN model, while the validation groups were used to verify the model's performance.
In validation Group-I, the model achieved an area under the curve (AUC) of 0.843. In validation Group-II, the kappa values for the model's prediction of calcium oxalate and uric acid components, consistent with infrared spectroscopy analysis, were 0.649 and 0.653, respectively.
Faster R-CNN demonstrates a robust capability for quantitative prediction of the composition of urinary calculi, indicating substantial promise for clinical applications.
探讨人工智能算法——基于区域的快速卷积神经网络（Faster R-CNN）从CT图像中识别尿路结石成分的能力及临床应用潜力。
本研究为回顾性研究。收集了2020年8月至2023年12月在河北大学附属医院接受治疗的776例尿路结石患者的数据。单纯结石患者按5:1的比例随机分为模型构建组和验证组-I，同时随机选取60例混合结石患者组成验证组-II。模型构建组用于构建和测试Faster R-CNN模型的性能，而验证组用于验证模型的性能。
在验证组-I中，模型的曲线下面积（AUC）为0.843。在验证组-II中，模型对草酸钙和尿酸成分预测的kappa值分别为0.649和0.653，与红外光谱分析结果一致。
Faster R-CNN在尿路结石成分的定量预测方面表现出强大的能力，具有显著的临床应用前景。
