Kidney Week

Abstract: TH-PO032

Deep Learning Tubulometrics to Stratify Tubular Injury in Whole-Slide Kidney Specimens

Session Information

• Augmented Intelligence for Prediction and Image Analysis
  October 24, 2024 | Location: Exhibit Hall, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Augmented Intelligence, Digital Health, and Data Science

• 300 Augmented Intelligence, Digital Health, and Data Science

Authors

• Klaus, Martin, Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Martin Klaus

• Gaßmann, Philipp M., Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Philipp M. Gaßmann

• Abdullah, Rosch, Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Rosch Abdullah

• Ehreiser, Louisa, Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Louisa Ehreiser

• Long, Hao, Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Hao Long, MD

• Lichtnekert, Julia, Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Julia Lichtnekert, MD

• Steiger, Stefanie, Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Stefanie Steiger, PhD

• Lech, Maciej, Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Maciej Lech, PhD

• Anders, Hans J., Division of Nephrology, Department of Medicine IV, University Hospital, Ludwig-Maximilians-Universität München, Munich, Germany

Hans J. Anders, MD

Background

Acute kidney injury is highly prevalent and most often involves tubular injury. However, studies on detailed assessment of tubular epithelial cell morphometry within tubuli are missing. To this end, we provide a deep learning-enhanced method to segment and analyze tubular epithelial cells and tubuli ("tubulometrics") in whole slide kidney specimens of wild-type mice and mice models of acute and chronic tubular injury.

Methods

Kidneys were embedded in paraffin and stained with PAS and immunohistochemistry before digital scanning. More than 1200 tubuli and 6800 tubular epithelial cells of control mice, mice suffering ischemia-reperfusion injury or chronic tubular injury were manually annotated. Subsequently, a deep learning algorithm was trained to segment tubuli and tubular epithelial cells and morphometric readouts were obtained. Furthermore, tubulometrics were compared to traditional tubular injury scores.

Results

Tubular area, tubular cell number and density per tubulus section were identified as morphometric readouts. These tubulometrics were able to complement a traditional tubular injury score (PAS score) and interstitial fibrosis tubular atrophy score (IFTA). In wild-type mice PAS score was 0, IFTA 0, mean tubular area 2339 [µm²], mean tubular cell count per tubulus 7.27, and mean tubular cell density 3.09 [10^-3/µm²]. In mice with ischemia-reperfusion injury PAS score was 3.44, IFTA 10%, mean tubular area 1818 [µm²], mean tubular cell count per tubulus 2.02, mean tubular cell density 1.11 [10^-3/µm²]. In mice with crystal-induced chronic tubular injury PAS score was 3.06, IFTA 40%, mean tubular area 2746 [µm²], mean tubular cell count per tubulus 7.71, and mean tubular cell density 2.81 [10^-3/µm²]. The deep learning segmentation algorithm ensured time-efficiency and scalability by taking 5 minutes per slide.

Conclusion

We propose the analysis of tubular epithelial cell morphometry in kidney specimens to better identify and stratify tubular injury. Whereas tubular epithelial cell loss was found in acute injury, tubular hypertrophy together with cell regeneration were observed in the chronic model. Integrated with existing histopathological scores a more detailed assessment of kidney biopsies might become possible. Further studies are needed to validate the findings in larger cohorts.

Funding

• Government Support – Non-U.S.