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 <a class="link-external link-https" href="https://github.com/DeepGAInS/GAInS" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve the problems of intersection, contact and overlap in biomedical image instance segmentation. Specifically, current methods usually handle these challenges by individual modeling, ignoring the intrinsic correlations between these conditions. The authors propose a new method - **Gradient Anomaly - aware Biomedical Instance Segmentation (GAInS)** to utilize instance gradient information to perceive local gradient - abnormal regions, thereby modeling the spatial relationships between instances and refining the segmentation of local regions. ### Detailed interpretation 1. **Background and problem description**: - Instance segmentation plays a crucial role in biomedical image analysis, involving the precise identification and partitioning of biomedical objects or anatomical structures (such as chromosomes, cells, glands, etc.). - Biomedical instances are often densely clustered, resulting in contact, overlap and intersection problems, which pose challenges to instance segmentation. - These problems stem from factors such as dimension compression during the imaging process, and the adhesion and flattening of instances during the sample preparation process. - Existing methods usually adopt a scheme of detection followed by segmentation, but they are not effective in handling contact and overlap problems. 2. **Limitations of existing methods**: - Most of the existing methods are solutions for specific application scenarios and lack a unified framework to comprehensively handle intersection, contact and overlap problems. - For example, DoNet is used for overlapping cell segmentation, and TopoSeg and HoverNet use topological and orientation information for nuclear segmentation in pathological images, but these methods are too specific and cannot be widely applied. 3. **Proposed new method**: - **GAInS**: This method uses the gradient field to capture distance and orientation information as effective guidance to model spatial information, enhancing the model's ability to capture and represent instance contact and overlap regions. - **Gradient Anomaly Mapping Module (GAMM)**: Encodes the radial field of instances and obtains instance gradient - abnormal maps through a sliding window. - **Adaptive Local Refinement Module (ALRM)**: Introduces a gradient - anomaly - aware loss function to efficiently refine boundaries and regions with gradient - anomaly attention. 4. **Innovation points**: - GAInS provides a new perspective to handle complex biomedical instance segmentation problems by combining gradient - abnormal information. - The model can not only capture the spatial relationships between instances, but also improve the segmentation accuracy through the adaptive refinement module. 5. **Experimental results**: - Extensive comparison and ablation experiments in three biomedical scenarios show that GAInS outperforms other state - of - the - art instance segmentation methods. - Specifically, GAInS has achieved the highest mAP scores on the ISBI2014, UCOC and Kaggle2018 datasets, and has a significant improvement in metrics such as AJI. ### Conclusion GAInS provides a unified and general method to handle the problems of intersection, contact and overlap in biomedical instance segmentation by introducing a gradient - anomaly - aware mechanism. The experimental results prove the effectiveness and superiority of this method, showing its broad application prospects in complex biomedical image analysis.

GAInS: Gradient Anomaly-aware Biomedical Instance Segmentation

IRLSG: Invariant Representation Learning for Single-Domain Generalization in Medical Image Segmentation

Biological Instance Segmentation with a Superpixel-Guided Graph.

GradiSeg: Gradient-Guided Gaussian Segmentation with Enhanced 3D Boundary Precision

GS-EMA: Integrating Gradient Surgery Exponential Moving Average with Boundary-Aware Contrastive Learning for Enhanced Domain Generalization in Aneurysm Segmentation

Boundary-Aware Gradient Operator Network for Medical Image Segmentation

Generative Medical Segmentation

GSAL: Geometric structure adversarial learning for robust medical image segmentation

GL-Segnet: Global-Local representation learning net for medical image segmentation

SegAN: Adversarial Network with Multi-scale $L_1$ Loss for Medical Image Segmentation

Biologically-inspired Semi-supervised Semantic Segmentation for Biomedical Imaging

Mirrored X-Net: Joint Classification and Contrastive Learning for Weakly Supervised GA Segmentation in SD-OCT★

Anomaly segmentation in retinal images with poisson-blending data augmentation

Object-Guided Instance Segmentation With Auxiliary Feature Refinement for Biological Images

Quantitative susceptibility mapping based basal ganglia segmentation via AGSeg: leveraging active gradient guiding mechanism in deep learning

Hybrid Intelligent-Annotation Organ Segmentation on Medical Datasets

Gradient Alignment Improves Test-Time Adaptation for Medical Image Segmentation

Look in Different Views: Multi-Scheme Regression Guided Cell Instance Segmentation

GlanceSeg: Real-time microaneurysm lesion segmentation with gaze-map-guided foundation model for early detection of diabetic retinopathy

Key Information-Guided Networks for Medical Image Segmentation in Medical Systems

ASC-Net: Unsupervised Medical Anomaly Segmentation Using an Adversarial-based Selective Cutting Network