Measurement of ureteral length: Comparison of deep learning-based method and other estimation methods on CT and KUB

• Published in: Computers in biology and medicine Vol. 184
• Main Authors: Wang, Kexin, Zhao, Zheng, Liu, Yi, Nai, Rile, Yuan, Changwei, Wu, Pengsheng, Li, Jialun, Zhang, Xiaodong, Wang, He
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2025
• Subjects: CT urography; Deep learning; KUB radiography; Ureteral length; Deep learning; CT urography; Ureteral length; KUB radiography
• ISSN: 0010-4825
• DOI: 10.1016/j.compbiomed.2024.109374

Accurate preoperative assessment of ureteral length is crucial for effective ureteral stenting. Utilize a deep learning approach to measure ureter length on CT urography (CTU) images and compare the obtained results with those derived from other estimation methods. In a retrospective cohort (cohort A, n = 411), CTU images were collected and used to develop a 3D deep learning model for the segmentation of bilateral ureters. The centerline of the ureters was determined based on the segmentation, and the length of the ureters was automatically obtained (CTU_ai). Another cohort (cohort B, n = 220) was collected as the hold-out test for the model. All patients in cohort B had KUB, non-contrast enhanced CT (CT NoC), and CTU images. Cohort B utilized eight measurement methods, with one annotated by two radiologists serving as the reference standard (CTU_ref) and the remaining seven as the studied methods, including three measurement methods applied to CTU (CTU_ai, CTU_oblique, CTU_slice), two applied to CT NoC (CT_oblique, CT_slice), and two applied to KUB (KUB_short, KUB_long). The results of the seven studied methods were compared to those of the reference in cohort B. Among the 220 patients (96 females, 124 males), 437 ureters were measured for length (218 left, 219 right), with a median length of 24.7 (IQR 23.2–26.2) cm. No significant differences were observed between genders or laterality (both P > 0.05). Moreover, there was no correlation between ureteral length and age (r = −0.027, P = 0.573). The ureteral length measured by CTU_ai was not significantly different from that measured by CTU_ref (P = 0.514), whereas the length measured by the other studied methods was significantly different from that measured by CTU_ref (all P < 0.001). The ICC values with their 95 % confidence intervals (CIs) for the comparison between the reference standard (CTU_ref) and the other measurement methods: CTU_ai (ICC = 0.852, 95 % CI 0.825–0.876), CTU_oblique (ICC = 0.351, 95 % CI -0.083-0.689), CTU_slice (ICC = 0.269, 95 % CI -0.095-0.573), CTU_oblique_slice (ICC = 0.059, 95 % CI -0.032-0.218), CTU_slice (ICC = 0.049, 95 % CI -0.028-0.188), KUB_short (ICC = 0.151, 95 % CI 0.051–0.247), and KUB_long (ICC = 0.147, 95 % CI 0.034–0.253). For CTU_ai, in 89.0 % of the ureters, the ureteral length deviation was within 20 mm of the reference standard, which was the highest among all the studied methods (all P < 0.001). The deep learning model offers a reliable and accurate tool for ureteral length measurement on CTU images, which could enhance the effectiveness of ureteral stenting procedures. Its performance surpasses traditional measurement methods, making it a promising technology for integration into clinical practice. •A deep learning model based on CTU images was used to measure ureteral length and performed superiorly to other methods.•Bland-Altman analysis found a -1.3 mm bias between the model and reference, not significant (P = 0.057).•The model's differences from the reference standard didn't significantly differ by sex (P = 0.319) or laterality (P = 0.287).•The model showed the utmost consistency with the reference standard, with 89.0 % of ureters showing a deviation within 20 mm.