Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GAInS: Gradient Anomaly-aware Biomedical Instance Segmentation

Runsheng Liu, Hao Jiang, Yanning Zhou, Huangjing Lin, Liansheng Wang, Hao Chen

TL;DR

A Gradient Anomaly-aware Biomedical Instance Segmentation approach (GAInS), which leverages instance gradient information to perceive local gradient anomaly regions, thus modeling the spatial relationship between instances and refining local region segmentation.

Abstract

Instance segmentation plays a vital role in the morphological quantification of biomedical entities such as tissues and cells, enabling precise identification and delineation of different structures. Current methods often address the challenges of touching, overlapping or crossing instances through individual modeling, while neglecting the intrinsic interrelation between these conditions. In this work, we propose a Gradient Anomaly-aware Biomedical Instance Segmentation approach (GAInS), which leverages instance gradient information to perceive local gradient anomaly regions, thus modeling the spatial relationship between instances and refining local region segmentation. Specifically, GAInS is firstly built on a Gradient Anomaly Mapping Module (GAMM), which encodes the radial fields of instances through window sliding to obtain instance gradient anomaly maps. To efficiently refine boundaries and regions with gradient anomaly attention, we propose an Adaptive Local Refinement Module (ALRM) with a gradient anomaly-aware loss function. Extensive comparisons and ablation experiments in three biomedical scenarios demonstrate that our proposed GAInS outperforms other state-of-the-art (SOTA) instance segmentation methods. The code is available at https://github.com/DeepGAInS/GAInS.

GAInS: Gradient Anomaly-aware Biomedical Instance Segmentation

TL;DR

A Gradient Anomaly-aware Biomedical Instance Segmentation approach (GAInS), which leverages instance gradient information to perceive local gradient anomaly regions, thus modeling the spatial relationship between instances and refining local region segmentation.

Abstract

Instance segmentation plays a vital role in the morphological quantification of biomedical entities such as tissues and cells, enabling precise identification and delineation of different structures. Current methods often address the challenges of touching, overlapping or crossing instances through individual modeling, while neglecting the intrinsic interrelation between these conditions. In this work, we propose a Gradient Anomaly-aware Biomedical Instance Segmentation approach (GAInS), which leverages instance gradient information to perceive local gradient anomaly regions, thus modeling the spatial relationship between instances and refining local region segmentation. Specifically, GAInS is firstly built on a Gradient Anomaly Mapping Module (GAMM), which encodes the radial fields of instances through window sliding to obtain instance gradient anomaly maps. To efficiently refine boundaries and regions with gradient anomaly attention, we propose an Adaptive Local Refinement Module (ALRM) with a gradient anomaly-aware loss function. Extensive comparisons and ablation experiments in three biomedical scenarios demonstrate that our proposed GAInS outperforms other state-of-the-art (SOTA) instance segmentation methods. The code is available at https://github.com/DeepGAInS/GAInS.
Paper Structure (11 sections, 6 equations, 4 figures, 3 tables)

This paper contains 11 sections, 6 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Gradient Anomaly Modeling
  4. Adaptive Local Refinement Module
  5. Special Cases of Gradient Anomaly Generation
  6. Experiments and Results
  7. Dataset and Experiments Settings
  8. Ablation Studies and Analysis
  9. Comparison with SOTA methods
  10. Conclusion
  11. Acknowledgments

Figures (4)

  • Figure 1: The schematic illustration of biomedical instance segmentation: (a) medical images with overlapping cells, crossing chromosomes, and touching nuclei; (b) zoomed insets of local regions; (c) the corresponding ground truth masks; (d) gradient anomaly fields; (e) instance-level gradient anomaly maps.
  • Figure 2: Overview of the proposed GAInS. The network contains two main prediction branches: (1) gradient branch with GAMM for gradient anomaly learning and gradient anomaly field mapping; (2) mask branch with ALRM to adaptively refine CTO regions, such as the heavily overlapping regions of the cervical cells illustrated.
  • Figure 3: The workflow of gradient anomaly map generation. (a) original images with (b) GT masks $M^{m}$ of interacting instances ($Ins1$, $Ins2$, $Ins3$); (c) radial distance map $R_{s}$; (d) radial direction map $R_{r}$; (e) top view $M_{W}$; (f) window sliding with deviation assignment; (g) gradient anomaly map $M^{g}$, and (h) combined image.
  • Figure 4: Qualitative results of our GAInS and other SOTA methods. The red boxes highlight the CTO regions of cell, chromosome and nucleus.