Abstract:Deep learning models have shown promising performance for cell nucleus segmentation in the field of pathology image analysis. However, training a robust model from multiple domains remains a great challenge for cell nucleus segmentation. Additionally, the shortcomings of background noise, highly overlapping between cell nucleus, and blurred edges often lead to poor performance. To address these challenges, we propose a novel framework termed CausalCellSegmenter, which combines Causal Inference Module (CIM) with Diversified Aggregation Convolution (DAC) techniques. The DAC module is designed which incorporates diverse downsampling features through a simple, parameter-free attention module (SimAM), aiming to overcome the problems of false-positive identification and edge blurring. Furthermore, we introduce CIM to leverage sample weighting by directly removing the spurious correlations between features for every input sample and concentrating more on the correlation between features and labels. Extensive experiments on the MoNuSeg-2018 dataset achieves promising results, outperforming other state-of-the-art methods, where the mIoU and DSC scores growing by 3.6% and 2.65%.

What problem does this paper attempt to address?

The paper primarily addresses the challenges faced in the task of nuclear segmentation in pathological image analysis, including background noise, highly overlapping nuclei, and blurred edges, by proposing a new framework called CausalCellSegmenter. This framework combines a Causal Inference Module (CIM) with Diversified Aggregation Convolution (DAC) technology. Specifically, the problems and solutions mentioned in the paper are as follows: 1. **Background noise, highly overlapping nuclei, and blurred edges**: These issues limit the performance of existing methods in nuclear segmentation. - **Solution**: By introducing the DAC module to optimize the feature extraction process, this module can integrate multi-scale feature maps and use a simple parameter-free attention mechanism (SimAM) to enhance the feature representation of regions of interest, thereby improving the distinction between nuclei and non-nuclei. 2. **Data distribution differences**: Due to significant variations in illumination, resolution, and staining techniques in pathological images obtained from different hospitals or equipment, it is challenging to achieve ideal segmentation performance when applying deep learning models to clinical practice. - **Solution**: By introducing the CIM to alleviate this issue. This module adopts a sample weight adjustment strategy to directly remove spurious correlations between the features of each input sample, making the model more focused on the correlation between features and labels. The key contributions of the paper include: 1. The design of the **DAC module** to optimize the fusion of different types of semantic information, reducing the occurrence of false positive predictions. 2. The introduction of **CIM**, which uses a causal inference learning strategy to dynamically adjust sample weights, enhancing the recognition ability of nuclear features and effectively reducing the negative impact of spurious correlations. Through extensive experiments on the MoNuSeg-2018 dataset, CausalCellSegmenter achieved significantly better results than other advanced methods, with improvements of 3.6% and 2.65% in mIoU and DSC metrics, respectively, demonstrating its effectiveness in addressing the nuclear segmentation task.

CausalCellSegmenter: Causal Inference inspired Diversified Aggregation Convolution for Pathology Image Segmentation

Mutual-Complementing Framework for Nuclei Detection and Segmentation in Pathology Image

DSCA-Net: Double-stage Codec Attention Network for automatic nuclear segmentation

CellSAM: Advancing Pathologic Image Cell Segmentation via Asymmetric Large‐Scale Vision Model Feature Distillation Aggregation Network

An Accurate Nuclei Segmentation Algorithm in Pathological Image Based on Deep Semantic Network

CellSegNet: an Adaptive Multi-Resolution Hybrid Network for Cell Segmentation

Accurate Segmentation of Nuclei in Pathological Images Via Sparse Reconstruction and Deep Convolutional Networks

Multi-channel pathological image segmentation with gated axial self-attention

MCCSeg: Morphological embedding causal constraint network for medical image segmentation

Advanced Multi-Microscopic Views Cell Semi-supervised Segmentation

ClusterSeg: A Crowd Cluster Pinpointed Nucleus Segmentation Framework with Cross-Modality Datasets

Deep Adversarial Training for Multi-Organ Nuclei Segmentation in Histopathology Images

Black-box Domain Adaptative Cell Segmentation Via Multi-source Distillation

Multi-task Generative Adversarial Learning for Nuclei Segmentation with Dual Attention and Recurrent Convolution

A pathology image segmentation framework based on deblurring and region proxy in medical decision-making system

Dual U-Net for the Segmentation of Overlapping Glioma Nuclei.

A Dual-Path Instance Segmentation Network Based on Nuclei Contour in Histology Image

Attentional Dilated Convolution Neural Network for Nuclei Segmentation in Histopathology Images

Efficient Nucleus Detection in Digital Pathology Images Using Multi-task and Multi-scale Instance Segmentation Network

The Multi-modality Cell Segmentation Challenge: Towards Universal Solutions

CIA-Net: Robust Nuclei Instance Segmentation with Contour-aware Information Aggregation