ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on X

Kidney Week

Abstract: TH-PO024

Automated Segmentation for Measuring Kidney Volume and Length in CT Images: A Cross-National Evaluation Study

Session Information

Category: Augmented Intelligence, Digital Health, and Data Science

  • 300 Augmented Intelligence, Digital Health, and Data Science

Authors

  • Zhang, Zhongyi, The Second Hospital of Shandong University, Jinan, Shandong, China
  • Hui, Yiyi, Shandong Provincial Qianfoshan Hospital, Jinan, Shandong, China
  • Liu, Gang, The Second Hospital of Shandong University, Jinan, Shandong, China
  • Sun, Lina, Linyi People's Hospital, Linyi, Shandong, China
  • Liu, Xiangchun, The Second Hospital of Shandong University, Jinan, Shandong, China
Background

Kidney structure is an essential biomarker for evaluating kidney diseases. The main kidney structure biomarkers include the total kidney volume (TKV), parenchymal kidney volume (PKV), and kidney lengths. Although the most common form of diagnostic kidney imaging is ultrasound in a 2D manner, in more complex clinical settings or emergencies, patients usually only have CT scans for other diseases, such as acute abdominal pain, or abdominal surgery; so renal physicians need to estimate the kidney structure directly from CT imaging for patients who have kidney injury to differential diagnosis between CKD and acute kidney diseases. Measuring these biomarkers is labor-intensive for radiologists to require manual declinations of kidney boundaries slice by slice. However, an automated method of kidney structure measurements in both non-contrast and contrast CT images is still not established.

Methods

Our retrospective study included a total of 1,153 CT images with and without contrast materials from nine countries. Reference standards of kidney volumes and lengths were manually measured and validated by one experienced radiologist. We developed a deep-learning (DL) method for measuring the TKV, PKV, and kidney lengths in CT images. The DL model was trained and tuned on the development data (n=526), and then the model was tested on a held-out internal test set (n=166) and external test sets (n=461). Kidney volumes and lengths were then automatically measured using 3D kidney segmentations.

Results

DL-based total kidney segmentation achieved an internal Dice similarity coefficient (DSC) of 0.986 ± 0.013 and an external DSC of 0.967 ± 0.016. In addition, parenchymal segmentation achieved an internal DSC of 0.977 ± 0.019 and an external DSC of 0.948 ± 0.020. DL-based measurements of TKV, RPV, and kidney lengths demonstrated statistical equivalence with the reference standards across all test images (p > 0.05).

Conclusion

We developed a fully automated segmentation method to measure TKV, PKV, and kidney lengths in CT images. The fully automated approach achieved a radiologist-equivalent accuracy on both contrast-enhanced and unenhanced CT modalities. With public access to algorithms and segmentations, our method could identify more potential kidney disease patients and reduce the clinical workload of renal physicians and imaging experts.