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: FR-PO043

Automated Detection of Renal Contrast Phases

Session Information

Category: Augmented Intelligence, Digital Health, and Data Science

  • 300 Augmented Intelligence, Digital Health, and Data Science

Authors

  • Rassoulinejad Mousavi, Seyed Moein, Mayo Foundation for Medical Education and Research, Rochester, Minnesota, United States
  • Khosravi, Bardia, Mayo Foundation for Medical Education and Research, Rochester, Minnesota, United States
  • Weston, Alexander, Mayo Foundation for Medical Education and Research, Rochester, Minnesota, United States
  • Garner, Hillary, Mayo Foundation for Medical Education and Research, Rochester, Minnesota, United States
  • Takahashi, Naoki, Mayo Foundation for Medical Education and Research, Rochester, Minnesota, United States
  • Kline, Timothy L., Mayo Foundation for Medical Education and Research, Rochester, Minnesota, United States
  • Erickson, Bradley J., Mayo Foundation for Medical Education and Research, Rochester, Minnesota, United States

Group or Team Name

  • Mayo AI Lab.
Background

Renal contrast phases are vital for evaluating renal function, assessing pathology, detecting abnormalities, and formulating treatment plans. In this study, we utilized a combination of deep learning (DL) and regression techniques to identify contrast agents in CT scans and determine the specific stages of contrast-enhanced (CE) renal imaging. Initially, DL was employed to differentiate between CE and non-CE scans. For scans with administered contrast agents, a random forest regression model was trained to predict the complete range of values associated with the contrast phases. This approach allows for a more precise analysis of the continuous chronological sequence of contrast phases, rather than relying on predefined categories like classification.

Methods

The DL model is trained using a ConvNeXt-Femto architecture to classify CE and non-CE renal imaging. We used 3033 CT scans from 1017 patients with renal cell cancer. Using a segmentation model the left and right kidney were segmented. We selected five 2D slices of renals for classification: the middle slice based on the right kidney, two slices above, and two slices below. This simplified the input data while leveraging renal image characteristics. Features extracted from DL were used as input for a regression task using random forest to associate a value with each contrast phase based on chronological sequence aspects of renal enhancement. We employed five-fold cross-validation for training, using a sixth fold as the test set.

Results

The models performed well both in contrast detection and phase association tasks. The DL achieved an accuracy of 98% in classifying CE versus non-CE. For predicting other contrast phases, the model had a mean absolute error (MAE) of 0.34 on the test set. The model effectively associates numerical values with the eight contrast phases from "early corticomedullary" to "late pyelographic", enabling characterization of renal contrast enhancement patterns.

Conclusion

Coupling DL and regression, proved to be highly effective in automating the detection of contrast agents in CT scans and accurately determining the specific stages of CE in renal imaging. This approach enables comprehensive analysis of the continuous chronological sequence of contrast phases, surpassing the limitations of predefined categories typically used in classification.

Funding

  • Other NIH Support