PrPSeg: Universal Proposition Learning for Panoramic Renal Pathology Segmentation

TL;DR

This research introduces a novel univer-sal proposition learning approach, called panoramic re-nal pathology segmentation (PrPSeg), designed to segment comprehensively panoramic structures within kidney by in-tegrating extensive knowledge of kidney anatomy.

Abstract

Understanding the anatomy of renal pathology is crucial for advancing disease diagnostics, treatment evaluation, and clinical research. The complex kidney system comprises various components across multiple levels, including regions (cortex, medulla), functional units (glomeruli, tubules), and cells (podocytes, mesangial cells in glomerulus). Prior studies have predominantly overlooked the intricate spatial interrelations among objects from clinical knowledge. In this research, we introduce a novel universal proposition learning approach, called panoramic renal pathology segmentation (PrPSeg), designed to segment comprehensively panoramic structures within kidney by integrating extensive knowledge of kidney anatomy. In this paper, we propose (1) the design of a comprehensive universal proposition matrix for renal pathology, facilitating the incorporation of classification and spatial relationships into the segmentation process; (2) a token-based dynamic head single network architecture, with the improvement of the partial label image segmentation and capability for future data enlargement; and (3) an anatomy loss function, quantifying the inter-object relationships across the kidney.

Figures (6)

• Figure 1: Knowledge transformation from kidney anatomy to computational modeling -- This figure demonstrates the transformation of intricate clinical anatomical relationships within the kidney into a structured computational matrix. (a) Pathologists examine histopathology following the kidney anatomy. (b) This study revisits such kidney anatomy with hierarchical semantic taxonomy. (c) The proposed PrPSeg method further mathematically abstracts the semantic taxonomy as a universal proposition matrix. This matrix serves as a foundation for our computational model, reflecting the complex interplay of anatomical elements in the kidney.
• Figure 2: Universal proposition matrix with anatomy loss -- This figure shows the key innovation of the proposed method. (a) Multi-scale (region-level, unit-level, and cell-level) hierarchical semantic taxonomy is presented. (b) The proposed PrPSeg mathematically models the semantic taxonomy as a universal proposition matrix, which delineates robust constraints and relationships between anatomical entities. (c) We further encode the universal proposition matrix as a novel anatomy loss function, designed to operationalize the affirmative and negatory relationships inherent in kidney anatomy.
• Figure 3: Token-based dynamic head network architecture -- This figure illustrates the architecture of the proposed PrPSeg method. It incorporates a residual U-Net backbone, augmented with class-aware and scale-aware tokens. These tokens are integrated into each block of the encoder, as well as the Global Average Pooling (GAP) block, ensuring a comprehensive understanding of both class and scale features. Such features are aggregated by a fusion block to adaptively generate the parameters for a single dynamic segmentation head. The proposed method is able to segment all hierarchical semantic anatomies using a single network.
• Figure 4: Token-based dynamic head -- This figure visualizes the architecture of our proposed token-based dynamic head backbone. Central to our design is the ability to maintain a consistent model architecture while dynamically accommodating an increasing number of segmentation classes. This flexibility is achieved by extending the dimensions of the tokens, rather than altering the backbone structure. Key components include a dynamic token bank with class-aware and scale-aware tokens, an encoder, and a dynamic head network, all orchestrated to efficiently handle class expansion without necessitating changes to the backbone.
• Figure 5: Validation qualitative results -- This figure shows the qualitative results of different approaches. The proposed method achieved superior panoramic renal pathology segmentation on 8 classes range regions to cells with fewer false positives, false negatives, and morphological errors.