Evaluating Gates’ GFR accuracy in renal tumor patients through integration of standalone CT-derived renal depth estimation via deep learning segmentation: a preliminary study
- 1Department of Nuclear Medicine, The Third Affiliated Hospital of Sun Yat-sen University
Abstract
Background Gates’ Glomerular Filtration Rate (GFR) measurement via SPECT dynamic renal scintigraphy is common in clinical practice. However, its accuracy is limited by the variability of renal depth (RD) estimated by the conventional height- and weight-based Tonnesen formula, particularly in patients with renal tumors. This study compared a novel standalone CT RD estimation via deep learning segmentation and GFR measurement method against the Tonnesen formula between renal tumor and non-tumor groups, using the double plasma sample method (DPSM) as the reference standard. Methods A retrospective analysis was conducted on 99 patients from January to December 2024, categorized into renal tumor group (n = 23) and non-tumor group (n = 76). All patients underwent 99mTc-DTPA SPECT/CT dynamic renal scintigraphy, standalone abdominal CT, serum creatinine (Scr) testing, and DPSM. Five GFR measurement methods were compared and analyzed : (1) Gates’ method with deep learning-based standalone CT segmentation and RD estimation, (2) Gates’ method using Tonnesen’s RD via Siemens software, (3) Gates’ method using Tonnesen’s RD via MMIS software, (4) Scr-based estimation, and (5) DPSM serving as the reference standard. Welch’s independent t-tests were conducted statistically to compare various GFRs between tumor and non-tumor groups. Pearson correlation coefficients were used to evaluate relationships among the various GFR methods. Results Standalone CT-derived RDs were greater than Tonnesen formula estimation. Particularly in the tumor group, CT-derived vs. Tonnesen RDs were 7.30 ± 1.51 cm vs. 5.97 ± 1.07 cm (left), and 7.15 ± 1.27 cm vs. 6.01 ± 1.08 cm (right). The mean total GFR measurements using the reference standard DPSM, CT-derived RD, Tonnesen’s RD via Siemens and MMIS software, and Scr-based, were 87.91 ± 19.53, 94.61 ± 28.11, 77.29 ± 17.93, 79.28 ± 23.98, and 81.71 ± 23.31 mL/min for the tumor group, 77.56 ± 11.04, 81.30 ± 30.67, 68.07 ± 22.86, 68.22 ± 24.16, and 76.15 ± 30.65 mL/min for the non-tumor group. The CT-based measurements showed numerically higher correlation with DPSM (r = 0.915 for the tumor group; r = 0.825 for the non-tumor group), compared with Tonnesen-based measurements (r < 0.8 for both subgroups). In paired method-comparison analysis, the CT-based GFR calculation showed descriptively smaller absolute differences from DPSM (21.47 mL/min) than the Tonnesen-based GFR methods (Siemens: 32.57 mL/min, MMIS: 27.04 mL/min) in tumor group. Conclusions In this preliminary cohort, Gates’ GFR calculation incorporating standalone abdominal CT-derived RD estimation via deep learning-based segmentation was associated with a higher correlation with DPSM and descriptively smaller absolute differences from DPSM than conventional Gates GFR with Tonnesen-derived RD estimation in patients with renal tumor. By leveraging existing diagnostic CT imaging, this method provided an exploratory, patient-specific RD estimation strategy and GFR calculation without requiring an additional CT acquisition, but its reliability and clinical utility warrant further validation in a larger cohort.
Related
- Evaluating the real-world accessibility of open-source deep learning models for retinal vessel segmentation in optical coherence tomography angiography
- The integration of artificial intelligence into computed tomography: from segmentation to the standardisation of clinical decisions
- Foundation model-driven framework for human–object interaction prediction with segmentation mask integration
- Three-Dimensional Segmentation of Pulp Stones in Tooth-Level Volumes on Cone-Beam Computed Tomography Using nnU-Net v2: A Preliminary Study.
- Efficient and real-time polyp segmentation: An empirical evaluation of lightweight Mamba integration strategies
BibTeX
@article{wu2026evaluating,
title = {{Evaluating Gates’ GFR accuracy in renal tumor patients through integration of standalone CT-derived renal depth estimation via deep learning segmentation: a preliminary study}},
author = {Chunxing Wu and Guangfeng Chen and Qing Liao and Feng Zhang and Liangjun Xie and Yueming Zha},
journal = {BMC Nephrology},
year = {2026},
month = jul,
doi = {10.1186/s12882-026-05226-z},
url = {https://doi.org/10.1186/s12882-026-05226-z},
note = {CC BY},
}
RIS
TY - JOUR
TI - Evaluating Gates’ GFR accuracy in renal tumor patients through integration of standalone CT-derived renal depth estimation via deep learning segmentation: a preliminary study
AU - Chunxing Wu
AU - Guangfeng Chen
AU - Qing Liao
AU - Feng Zhang
AU - Liangjun Xie
AU - Yueming Zha
PY - 2026
DA - 2026/07/01
JO - BMC Nephrology
DO - 10.1186/s12882-026-05226-z
UR - https://doi.org/10.1186/s12882-026-05226-z
AB - Background Gates’ Glomerular Filtration Rate (GFR) measurement via SPECT dynamic renal scintigraphy is common in clinical practice. However, its accuracy is limited by the variability of renal depth (RD) estimated by the conventional height- and weight-based Tonnesen formula, particularly in patients with renal tumors. This study compared a novel standalone CT RD estimation via deep learning segmentation and GFR measurement method against the Tonnesen formula between renal tumor and non-tumor groups, using the double plasma sample method (DPSM) as the reference standard. Methods A retrospective analysis was conducted on 99 patients from January to December 2024, categorized into renal tumor group (n = 23) and non-tumor group (n = 76). All patients underwent 99mTc-DTPA SPECT/CT dynamic renal scintigraphy, standalone abdominal CT, serum creatinine (Scr) testing, and DPSM. Five GFR measurement methods were compared and analyzed : (1) Gates’ method with deep learning-based standalone CT segmentation and RD estimation, (2) Gates’ method using Tonnesen’s RD via Siemens software, (3) Gates’ method using Tonnesen’s RD via MMIS software, (4) Scr-based estimation, and (5) DPSM serving as the reference standard. Welch’s independent t-tests were conducted statistically to compare various GFRs between tumor and non-tumor groups. Pearson correlation coefficients were used to evaluate relationships among the various GFR methods. Results Standalone CT-derived RDs were greater than Tonnesen formula estimation. Particularly in the tumor group, CT-derived vs. Tonnesen RDs were 7.30 ± 1.51 cm vs. 5.97 ± 1.07 cm (left), and 7.15 ± 1.27 cm vs. 6.01 ± 1.08 cm (right). The mean total GFR measurements using the reference standard DPSM, CT-derived RD, Tonnesen’s RD via Siemens and MMIS software, and Scr-based, were 87.91 ± 19.53, 94.61 ± 28.11, 77.29 ± 17.93, 79.28 ± 23.98, and 81.71 ± 23.31 mL/min for the tumor group, 77.56 ± 11.04, 81.30 ± 30.67, 68.07 ± 22.86, 68.22 ± 24.16, and 76.15 ± 30.65 mL/min for the non-tumor group. The CT-based measurements showed numerically higher correlation with DPSM (r = 0.915 for the tumor group; r = 0.825 for the non-tumor group), compared with Tonnesen-based measurements (r < 0.8 for both subgroups). In paired method-comparison analysis, the CT-based GFR calculation showed descriptively smaller absolute differences from DPSM (21.47 mL/min) than the Tonnesen-based GFR methods (Siemens: 32.57 mL/min, MMIS: 27.04 mL/min) in tumor group. Conclusions In this preliminary cohort, Gates’ GFR calculation incorporating standalone abdominal CT-derived RD estimation via deep learning-based segmentation was associated with a higher correlation with DPSM and descriptively smaller absolute differences from DPSM than conventional Gates GFR with Tonnesen-derived RD estimation in patients with renal tumor. By leveraging existing diagnostic CT imaging, this method provided an exploratory, patient-specific RD estimation strategy and GFR calculation without requiring an additional CT acquisition, but its reliability and clinical utility warrant further validation in a larger cohort.
SN - 1471-2369
ER -