Abstract:Medical image segmentation is crucial in the field of medical imaging, aiding in disease diagnosis and surgical planning. Most established segmentation methods rely on supervised deep learning, in which clean and precise labels are essential for supervision and significantly impact the performance of models. However, manually delineated labels often contain noise, such as missing labels and inaccurate boundary delineation, which can hinder networks from correctly modeling target characteristics. In this paper, we propose a deep self-cleansing segmentation framework that can preserve clean labels while cleansing noisy ones in the training phase. To achieve this, we devise a gaussian mixture model-based label filtering module that distinguishes noisy labels from clean labels. Additionally, we develop a label cleansing module to generate pseudo low-noise labels for identified noisy samples. The preserved clean labels and pseudo-labels are then used jointly to supervise the network. Validated on a clinical liver tumor dataset and a public cardiac diagnosis dataset, our method can effectively suppress the interference from noisy labels and achieve prominent segmentation performance.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the negative impact of label noise (noisy labels) on the performance of models in medical image segmentation. Specifically, the author focuses on how to improve the robustness and accuracy of medical image segmentation models in the case of inaccurate or noisy labels in the training data. #### Core problems: 1. **The influence of label noise**: Manually - annotated medical image labels often contain noise, such as missing labels and inaccurate boundary annotations. These noises will interfere with the learning process of deep - learning models, causing the models to be unable to correctly model the target features. 2. **Limitations of existing methods**: - **Image - level cleaning methods**: They can only reduce the influence of overall noisy samples, but cannot deal with local noisy areas. - **Pixel - level cleaning methods**: Although they can identify and correct specific noisy pixels, they cannot distinguish between clean and noisy samples and are prone to introducing new noise. #### Solutions: To solve the above problems, the author proposes a framework named "Deep Self - cleansing Framework", which combines the advantages of image - level and pixel - level cleaning methods and cleans noisy labels while retaining clean labels during the training process. #### Main contributions: 1. **Label Filtering Module at Image - level (LFM)**: Model the loss distribution of each sample based on the Gaussian Mixture Model (GMM) to distinguish between clean labels and noisy labels. 2. **Label Cleaning Module at Pixel - level (LCM)**: Generate pseudo - labels with low noise through pseudo - labels for supervising network training. 3. **Iterative cleaning mechanism**: During the training process, apply LFM and LCM periodically to gradually clean noisy labels and ensure the stability and controllability of training. #### Experimental verification: The author conducted experiments on the liver tumor CT data set and the public cardiac MRI data set to verify the effectiveness and robustness of this method under different noise levels. Through these innovations, this paper provides an effective solution that can achieve high - quality segmentation results in medical image data sets with noisy labels.

Deep Self-Cleansing for Medical Image Segmentation with Noisy Labels

Clean Label Disentangling for Medical Image Segmentation with Noisy Labels

Anti-Interference From Noisy Labels: Mean-Teacher-Assisted Confident Learning for Medical Image Segmentation

A Hybrid Robust-Learning Architecture for Medical Image Segmentation with Noisy Labels.

Pick-and-Learn: Automatic Quality Evaluation for Noisy-Labeled Image Segmentation

Distilling Effective Supervision for Robust Medical Image Segmentation with Noisy Labels

Pathological Image Segmentation with Noisy Labels

Cross-denoising Network against Corrupted Labels in Medical Image Segmentation with Domain Shift

Toward high-quality pseudo masks from noisy or weak annotations for robust medical image segmentation

Suppressing label noise in medical image classification using mixup attention and self-supervised learning

Label Filling via Mixed Supervision for Medical Image Segmentation from Noisy Annotations

Label-noise-tolerant medical image classification via self-attention and self-supervised learning

Weakly Supervised Segmentation by a Deep Geodesic Prior

Reliable Mutual Distillation for Medical Image Segmentation under Imperfect Annotations

Labelling with dynamics: A data-efficient learning paradigm for medical image segmentation

A NOVEL WEAKLY SUPERVISED FRAMEWORK BASED ON NOISY-LABEL LEARNING FOR MEDICAL IMAGE SEGMENTATION

A Survey of Label-Noise Deep Learning for Medical Image Analysis

Annotation-Efficient Learning for Medical Image Segmentation Based on Noisy Pseudo Labels and Adversarial Learning

Transformation Consistent Self-ensembling Model for Semi-supervised Medical Image Segmentation

Deep learning with noisy labels: Exploring techniques and remedies in medical image analysis

Noisy Supervised Deep Learning for Remote Sensing Image Segmentation Using Electronic Maps