Abstract: TH-PO805
Automated Foot Process Width (FPW) Measurements Using a Deep Learning (DL) Model Are Interchangeable with Stereology and Correlate with Kidney Functional and Structural Variables
Session Information
- Pathology and Lab Medicine - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Pathology and Lab Medicine
- 1800 Pathology and Lab Medicine
Authors
- Smerkous, David D., University of Washington, Seattle, Washington, United States
- Mauer, Michael, University of Minnesota Twin Cities, Minneapolis, Minnesota, United States
- Tøndel, Camilla, Universitetet i Bergen, Bergen, Hordaland, Norway
- Svarstad, Einar, Universitetet i Bergen, Bergen, Hordaland, Norway
- Nelson, Robert G., NIDDK, Phoenix, Arizona, United States
- Najafian, Behzad, University of Washington, Seattle, Washington, United States
Background
Increased FPW is a key measure of podocyte (PC) health and injury. Unbiased stereology, the current gold standard, is time consuming and not widely available. To address this, we developed a DL model for automated FPW measurement.
Methods
A U-Net variant model was trained to semantically segment the PC-glomerular basement membrane interface (PGBMI) and filtration slits. A cloud-based application was developed for users to upload systematic electron microscopy (EM) images, view and download the results. The model was applied to EM images of biopsies from Fabry disease (FD) patients (n=56), type 2 diabetes (T2D) (n=15), minimal change disease (n=10), and controls (n=17). The results were compared with unbiased stereology measurements by expert technologists (ET).
Results
DL and ET FPW measurements were correlated and not statistically different in all of the groups. Bland-Altman plot confirmed method interchangeability. FPW measurement time/biopsy was reduced from ~8 hours (ET) to <1 min (DL). In male (M) FD patients, both DL and ET measured FPW correlated directly with age, urine protein/creatinine (PCR), podocyte volume (VPC) and volume of FD inclusions per PC [V(Inc/PC)]; However, only DL FPW correlated inversely with PC numerical density. In female (F) FD patients, only DL FPW directly correlated with PCR, VPC and V(Inc/PC). In the T2D group, both DL and ET FPW correlated directly with fractional volume of mesangial matrix per glomerulus. However, only DL FPW directly correlated with urine albumin/creatinine, and fractional volume of mesangium per glomerulus, and VPC, and inversely with glomerular filtration surface density.
Conclusion
Our novel and validated DL model for EM FPW measurements can make this important biomarker widely accessible for research and clinical applications.
Funding
- NIDDK Support