Abstract: TH-OR43
Automated Atubular Glomeruli Detection Using 3D Glomerular Quantification Algorithms
Session Information
- Pathology of Kidney Diseases: Novel Mechanisms and Clinical Correlations
October 22, 2020 | Location: Simulive
Abstract Time: 05:00 PM - 07:00 PM
Category: Pathology and Lab Medicine
- 1601 Pathology and Lab Medicine: Basic
Authors
- Yang, Haichun, Vanderbilt University Medical Center, Nashville, Tennessee, United States
- Deng, Ruining, Vanderbilt University, Nashville, Tennessee, United States
- Huo, Yuankai, Vanderbilt University, Nashville, Tennessee, United States
- Fogo, Agnes B., Vanderbilt University Medical Center, Nashville, Tennessee, United States
Background
Atubular glomeruli are associated with decreased glomerular filtration rate and kidney disease progression. To identify atubular glomerulus requires serial section analysis, tracking individual glomeruli, and then determining whether each glomerulus has or does not have connection to the proximal tubule. This process is labor intensive and very time-consuming, limiting its use. We aimed to test feasibility of automatically detecting atubular glomeruli by using Multi-Object Association for Pathology in 3D (Map3D).
Methods
The Map3D was created including a glomerular detection algorithm, dual-path multi-object tracking algorithm, and pixel-wise large-scale glomerular association algorithm across routine serial sectioning with whole slide imaging (WSI). Atubular glomerular counting was done on 6 normal mouse kidneys, and 3 mice with diphtheria toxin (DT)-mediated proximal tubule-specific injury in mice with tubular cell expression of the DT receptor, and 4 mice with patchy tubulointerstitial fibrosis induced by folic acid (FA). Data from this automated approach was compared with standard manual assessment detailed above and correlated with functional and structural parameters.
Results
The Map3D substantially reduced the time needed for average atubular glomerular counting per sample (30 min Map3D vs. 30 hours human). Atubular glomerular size were smaller than normal glomeruli. The number of complete (i.e. from pole to pole) glomeruli assessed increased by 25.6% using GQuant-3D (86±8.8 per mouse) vs human counting (72±2.7 per mouse). GQuant-3D recognized 14.3±5.5 atubular glomeruli per mouse sample, slightly less (83%) than the 16.7±3.0 atubular glomeruli per mouse sample recognized by manual human assessment. The percentage of atubular glomeruli by GQuant-3D was increased in DT mice (9.7±2.15%) and FA mice (36.5±7.44%) compared with normal mice (6.0±0.91%). The percentage of atubular glomeruli, counted by either GQuant-3D or human, correlated with interstitial fibrosis (R2=0.49 or 0.61 respectively), but not with tubular injury marker, KIM-1 and N-GAL.
Conclusion
The Map-3D algorithms reduced time required for atubular glomeruli assessment, provided data correlating well with human manual-based assessment, and correlated well with relevant morphology data. This methodology can be extended to 3D glomerular phenotype analysis.
Funding
- NIDDK Support