Clinical Relevance of Computationally Derived Tubular Features: Spatial Relationships and the Development of Tubulointerstitial Scarring in MCD/FSGS

Abstract

Background: Visual scoring of tubular damage has limitations in capturing the full spectrum of structural changes and prognostic potential. We investigate if computationally quantified tubular features can enhance prognostication and reveal spatial relationships with interstitial fibrosis.

Methods: Deep-learning and image-processing-based segmentations were employed in N=254/266 PAS-WSIs from the NEPTUNE/CureGN datasets (135/153 focal segmental glomerulosclerosis and 119/113 minimal change disease) for: cortex, tubular lumen (TL), epithelium (TE), nuclei (TN), and basement membrane (TBM). N=104 pathomic features were extracted from these segmented tubular substructures and summarized at the patient level using summary statistics. The tubular features were quantified across the biopsy and in manually segmented regions of mature interstitial fibrosis and tubular atrophy (IFTA), pre-IFTA and non-IFTA in the NEPTUNE dataset. Minimum Redundancy Maximum Relevance was used in the NEPTUNE dataset to select features most associated with disease progression and proteinuria remission. Ridge-penalized Cox models evaluated their predictive discrimination compared to clinical/demographic data and visual-assessment. Models were evaluated in the CureGN dataset.

Results: N=9 features were predictive of disease progression and/or proteinuria remission. Models with tubular features had high prognostic accuracy in both NEPTUNE and CureGN datasets and increased prognostic accuracy for both outcomes (5.6%-7.7% and 1.6%-4.6% increase for disease progression and proteinuria remission, respectively) compared to conventional parameters alone in the NEPTUNE dataset. TBM thickness/area and TE simplification progressively increased from non- to pre- and mature IFTA.

Conclusions: Previously under-recognized, quantifiable, and clinically relevant tubular features in the kidney parenchyma can enhance understanding of mechanisms of disease progression and risk stratification.

Figure 1:

Overview of the workflow: a. Study cohorts: Depicts the study sample selection process, where all the patients and whole slide images (WSIs) were passed through each of the sample selection processes following the patient-level and slide-level exclusion criteria resulting in the inclusion of 520 patients from 2 cohorts and 520 WSIs in this study. b. Ground truth and algorithmic segmentations: (1) Reference tissue from the University of Michigan, (2) Cortical area was annotated and quality controlled (QCed) by study pathologists and (3) Utilization of both deep-learning (DL) and traditional image-processing algorithms to segment tubular primitives from the PAS WSIs. c. Quantitative feature extraction: Extraction of 99 tubule-level and 5 biopsy-level features from the segmentation results in b using image-processing algorithms. d. Statistical analysis: tubular level features were aggregated to patient-level features which, along with biopsy-level features, were being clustered into feature groups where feature groups were ranked by minimum redundancy - maximum relevance (MRMR) and top feature groups were selected by ridge regression, and predictive value was assessed on these top features and their associations with outcomes were examined both in NEPTUNE as training and internal validation and in CureGN as external validation.

Figure 2.

Comparison of top 9 features values between reference tissue and cortical subregions. A. Periodic Acid-Schiff (PAS)-stained Whole Slide Images (WSIs) of reference tissue from nephrectomies without interstitial fibrosis and tubular atrophy (IFTA) (grey boundary), NEPTUNE non-IFTA (blue boundary), pre-IFTA (green boundary), and mature IFTA (red boundary) regions. B-J: Violin plots illustrating the top 9 normalized feature values in reference tissue (gray violin), non-IFTA (blue violin), pre-IFTA (green violin), and mature-IFTA (red violin) regions. B: The ratio between the area of the tubular epithelium and the entire tubule is comparable across reference tissue and non-IFTA regions, although several tubules in non-IFTA regions have low values, likely reflecting acute tubular injury (simplification of the tubular epithelium). Lower values are also present in tubules in pre-IFTA (reflecting simplification of the tubular epithelium), and mature IFTA regions (reflecting simplification of the tubular epithelium in the presence of thick tubular basement membranes). C-E: The average (C) and maximum thickness (D) and area (E) of the tubular basement membranes are slightly higher and more homogeneous in reference tissue compared to non-IFTA regions, likely reflecting an overall older age for reference tissue and the presence of acute tubular injury in non-IFTA regions and increases progressively in pre-IFTA and mature IFTA. F-G: The area of tubular epithelium or tubular epithelium + lumen occupied by nuclei increases progressively from reference tissue and non-IFTA regions to pre- and mature- IFTA, indicating simplification of the tubular epithelium due to acute tubular injury, pre-atrophy, and full atrophy of tubules, respectively. H: The proportion between tubular basement membrane area and nuclear area per tubule are comparable between reference tissue and non-IFTA regions. As tubules become progressively pre- and fully atrophic the tubular basement membranes become thicker and the ratio between nuclear and tubular basement membranes’ area lower. I: Overall, the distance between the center of the tubular nuclei and the border of the lumen is greater in reference tissue compared to the 3 cortical subregions. The comparable values between non-IFTA, pre-IFTA, and mature IFTA regions reflects simplification of the tubular epithelium due to acute tubular injury, pre- and fully atrophic tubules, respectively. J: In mature IFTA, the inner boundary of the tubular basement membrane is more irregular compared to other cortical subregions and reference tissue. The 4 features (B, C, E and I) are the features which are still significant after adjustment for NEPTUNE.

Figure 3:

Illustrations of tubular phenotypes for the top 9 predictive features with the segmentation results: tubular epithelium in red, tubular lumen in black, tubular basement membrane in green, and tubular nuclei in white. (top panel) The initial five features are displayed in a sequence from left to right, illustrating a progression from lower to higher values of the feature. (middle panel) For the next four features, the sequence is arranged from left to right, depicting a gradient from higher to lower expression of the (lower panel) Cartoon illustrating the spectrum the change from normal to atrophy. A blue scale bar, located in the bottom right corner of each tubule image, represents a length of 40 micrometers.

Figure 4

Feature comparison for NEPTUNE at different age intervals. For each violin plot, from left to right, it represents feature values in non-IFTA regions for patients with age 0~10, 0~18, 11~18 and greater than 18 respectively.

Figure 5

NEPTUNE patient-level feature comparison. Left patient is one patient with disease progression while the right patient is the one without disease progression. The table below shows the raw feature values for the 4 features which are still significant after adjustment between these 2 patients.