Automatic Segmentation of The Renal Collecting System in 3D Pediatric Ultrasound to Assess the Severity of Hydronephrosis

摘要

Hydronephrosis is a common kidney abnormality in the pediatric population. It is generally defined as the dilation of the renal collecting system due to a build-up of urine. Ultrasound is a safe imaging modality to detect hydronephrosis. Hydronephrosis index (HI) is one of the metrics used for hydronephrosis severity assessment. In this paper, we first develop a novel automated collecting system segmentation method using a 3D U-net deep neural network. The initial segmentation is refined using anatomical location prior of the renal fat spots around the collecting system. Then, we measure HI to assess the severity of hydronephrosis. The performance of the method was evaluated using a dataset of 3D ultrasound images from 64 hydronephrotic cases, showing an average Dice similarity coefficient of 0.76 ± 0.12, an average symmetric surface distance of 1.29 ± 0.95 mm, and an average HI error value of 2.1 ± 2.8 %.