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

Methods: Deep-learning and image-analysis approaches were employed on 254/266 Periodic acid Schiff-stained whole slide image (WSI) kidney biopsies from participants in the NEPTUNE/CureGN prospective observational cohort studies (135/153 with focal segmental glomerulosclerosis (FSGS) and 119/113 with minimal change disease (MCD)) to segment cortex, tubular lumen (TL), epithelium (TE), nuclei (TN), and basement membrane (TBM). One hundred four pathomic features were extracted from these segmented tubular substructures and aggregated at the patient level using summary statistics. In the NEPTUNE dataset, tubular features were quantified at the WSI level and in manually segmented regions of mature interstitial fibrosis and tubular atrophy (IFTA), pre-IFTA, and non-IFTA. Minimum Redundancy Maximum Relevance was then used 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: Nine features were predictive of disease progression and/or proteinuria remission. Models with tubular features had high prognostic accuracy in both NEPTUNE and CureGN, and higher prognostic accuracy for both outcomes compared to conventional parameters alone in NEPTUNE. TBM thickness/area and TE flattening and/or reduced cell size progressively increased from non- to pre- and mature IFTA.

Conclusions: Previously underrecognized computationally derived and quantifiable tubular characteristics may contribute to improving prognostic accuracy and risk stratification in patients with FSGS/MCD. Future studies are needed to test their generalizability across different diseases and populations before they can be deployed in clinical practice.