Attention-Based Point Cloud Completion for CT-Equivalent Kidney Volumetry from Tracked Freehand Ultrasound.
Authors
Affiliations (8)
Affiliations (8)
- Department of Urology, Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, 210002, China.
- School of Mathematics and Physics, Xi'an Jiaotong-Liverpool University, Suzhou, China.
- School of Computer Science and Information Engineering, Changzhou Institute of Technology, Changzhou, China.
- School of Computer and Information Engineering, Nanjing Tech University, Nanjing, 211816, China.
- Carbon Medical Device Ltd., Shenzhen, 518000, China.
- School of Artificial Intelligence, Nanjing University of Information Science and Technology, Nanjing, China.
- School of Mathematics and Physics, Xi'an Jiaotong-Liverpool University, Suzhou, China. [email protected].
- Department of Urology, Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, 210002, China. [email protected].
Abstract
Kidney volume measurement is critical for managing polycystic kidney disease and monitoring transplants, but computed tomography involves radiation and conventional ultrasound has 20-30% error. This study validates tracked freehand ultrasound with attention-based point cloud completion for kidney volumetry, with the primary objective of determining equivalence to computed tomography within a ± 2% margin. Sixty healthy volunteers (30 male, 30 female; mean age 44 years; enrolled March-August 2024; 120 kidneys) undergoing routine health examination computed tomography underwent electromagnetic-tracked ultrasound using three standardized scanning maneuvers and same-day computed tomography. Sparse point clouds from ultrasound were completed using an attention-based transformer network (PointAttN) and four benchmark architectures. Volumes were compared against computed tomography using equivalence testing on per-kidney relative differences within a ± 2% margin. Tracked freehand ultrasound achieved mean absolute error of 3.94 mL (3.0% relative error) in multiaxis merged mode and passed statistical equivalence testing (TOST: p<sub>1</sub> < 0.001, p<sub>2</sub> = 0.043). Single-axis scanning failed equivalence testing across all methods with 2-4 times higher errors. Tracked freehand ultrasound with attention-based point cloud completion demonstrates kidney volume measurement accuracy approaching CT-level precision in healthy volunteers, passing formal statistical equivalence testing within a 2% margin. These algorithm-based results require validation in clinical populations with renal pathology and comparison with radiologist assessment before broader deployment. Multiaxis scanning is necessary for clinical accuracy.