ASN's Mission

ASN leads the fight to prevent, treat, and cure kidney diseases throughout the world by educating health professionals and scientists, advancing research and innovation, communicating new knowledge, and advocating for the highest quality care for patients.

learn more

Contact ASN

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

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: PO1566

Automated Instance Cyst Segmentation of Polycystic Kidneys in MRIs

Session Information

Category: Genetic Diseases of the Kidneys

  • 1001 Genetic Diseases of the Kidneys: Cystic

Authors

  • Gregory, Adriana, Mayo Clinic, Department of Nephrology and Hypertension, Rochester, Minnesota, United States
  • Anaam, Deema A., Mayo Clinic, Department of Radiology, Rochester, Minnesota, United States
  • Vercnocke, Andrew, Mayo Clinic, Department of Radiology, Rochester, Minnesota, United States
  • Edwards, Marie E., Mayo Clinic, Department of Radiology, Rochester, Minnesota, United States
  • Chebib, Fouad T., Mayo Clinic, Department of Nephrology and Hypertension, Rochester, Minnesota, United States
  • Torres, Vicente E., Mayo Clinic, Department of Nephrology and Hypertension, Rochester, Minnesota, United States
  • Harris, Peter C., Mayo Clinic, Department of Nephrology and Hypertension, Rochester, Minnesota, United States
  • Erickson, Bradley J., Mayo Clinic, Department of Radiology, Rochester, Minnesota, United States
  • Kline, Timothy L., Mayo Clinic, Department of Radiology, Rochester, Minnesota, United States
Background

Autosomal dominant polycystic kidney disease (ADPKD) is characterized by the development of multiple cysts in the kidneys. Currently, total kidney volume (TKV) is used as the only imaging biomarker to monitor disease progression. However, the ADPKD phenotype can vary widely among patients presenting with similar TKVs. In this study, we developed an MR image analysis method that automatically segments and differentiates individual cysts (i.e. instance-based segmentation) within the kidneys of patients affected by ADPKD.

Methods

A total of 60 T2-weighted MR images representative of different ADPKD stages and phenotypes were identified from our database (TKV range: 296mL-9690mL). The automated 3D instance cyst segmentation model was developed using a convolutional neural network. We reformulated the instance segmentation task by training the model to learn cyst edges and cores separately. The instance labeling was later generated by a combination of connected components and the watershed algorithm. The network was trained on 30 images and validated on 10 images using a 4-fold cross validation technique. The remaining 20 images were used for testing and were compared to manual tracings from two independent readers.

Results

An example of the automated method performance is shown in figure 1A. Quantification of the automatically generated cysts (Fig. 1B.) showed strong correlation with the number of cysts detected by readers 1 and 2 with an R2 of 0.96 and 0.88, respectively. The cystic index showed high correlation with an R2 of 0.92 and 0.90 for the comparisons between the automated method and readers 1 and 2, respectively.

Conclusion

We developed and tested the first fully automated instance cyst segmentation method for patients affected by ADPKD. The results demonstrate the feasibility and high accuracy of performing cyst counting and measuring total cyst volume and cystic index automatically.

Funding

  • NIDDK Support