Ip06-20 Comparison of 1 Mm Vs 0.5-mm Thick Reconstructed Images to Evaluate Residual Stone Fragments and Dust Particles: an Evaluation Using Porcine Kidneys
Year: 2025
Type: article
Abstract: METHODS: Preoperative CT images and clinical information were collected from patients who underwent kidney stone surgery at Beijing Tsinghua Changgung Hospital from 2021 to 2023.The data were divided into training and testing sets at a 7:3 ratio.Stone composition was determined using Fourier-transform infrared spectroscopy.A generative adversarial network (GAN) algorithm was employed for super-resolution reconstruction of the CT images.Radiomic features were extracted from the super-resolution-enhanced CT images.The radiomic features were combined with clinical features, and LASSO was used for feature selection.Various machine learning methods were employed for model construction, and their performance was compared.RESULTS: A total of 521 patients were included in this study, comprising 197 cases of calcium oxalate stones, 146 infection stones, 60 uric acid stones, and 15 cystine stones.From 108 raw features extracted from each sample, 13 key features were selected.Two important clinical features (urine culture results and age) were identified.The model constructed using SVM demonstrated the best classification performance, with a micro-average AUC of 0.92 and a macro-average AUC of 0.87.In terms of specific stone composition, the prediction for cystine stones showed the highest performance (AUC[0.99),followed by calcium oxalate stones (AUC[0.84),infection stones (AUC[0.79),and uric acid stones (AUC[0.82).CONCLUSIONS: The radiomics model based on non-contrast CT images can accurately predict the composition of various kidney stones preoperatively.
Source: The Journal of Urology
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4409268792