Abstract:Patients undergoing chest X-rays (CXR) often endure multiple lung diseases. When evaluating a patient's condition, due to the complex pathologies, subtle texture changes of different lung lesions in images, and patient condition differences, radiologists may make uncertain even when they have experienced long-term clinical training and professional guidance, which makes much noise in extracting disease labels based on CXR reports. In this paper, we re-extract disease labels from CXR reports to make them more realistic by considering disease severity and uncertainty in classification. Our contributions are as follows: 1. We re-extracted the disease labels with severity and uncertainty by a rule-based approach with keywords discussed with clinical experts. 2. To further improve the explainability of chest X-ray diagnosis, we designed a multi-relationship graph learning method with an expert uncertainty-aware loss function. 3. Our multi-relationship graph learning method can also interpret the disease classification results. Our experimental results show that models considering disease severity and uncertainty outperform previous state-of-the-art methods.

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are the label uncertainty existing in the current chest X - ray image classification datasets and the insufficient consideration of disease severity. Specifically: 1. **Label Uncertainty**: When extracting labels from the existing chest X - ray image classification datasets, the uncertainty expressed by radiologists in clinical reports is not considered. For example, radiologists may use words such as "probable" and "not excluded" to indicate the uncertainty of diagnosis, but this information is usually ignored or simply processed as a deterministic label (such as 0 - 1 coding) in the existing datasets. 2. **Disease Severity**: When extracting labels from the existing datasets, the severity of the disease is also not considered. For example, "mild effusion" and "severe cardiomegaly" may be processed as the same label in the existing datasets, but in fact their clinical significance and treatment plans may be very different. To solve these problems, the author proposes the following methods: - **Relabeling Diseases**: Through cooperation with clinical experts, use a rule - based method to relabel diseases from chest X - ray reports, taking into account the severity and uncertainty of diseases. - **Multi - Relational Graph Learning Method**: Design a multi - relational graph learning method, and improve the interpretability and classification performance of the model by introducing an expert - uncertainty - aware loss function. - **Interpret Disease Classification Results**: The proposed method can not only improve the classification accuracy, but also interpret the disease classification results, helping radiologists verify the accuracy of the model and build trust in the AI system. Through these improvements, the author hopes to construct a more realistic and comprehensive chest X - ray image classification dataset, thereby improving the performance and interpretability of the model.

Expert Uncertainty and Severity Aware Chest X-Ray Classification by Multi-Relationship Graph Learning

Multi-label local awareness and global co-occurrence priori learning improve chest X-ray classification

Multi-Label Learning With Visual-Semantic Embedded Knowledge Graph for Diagnosis of Radiology Imaging

Multi-label Chest X-ray Image Classification Via Category-Wise Residual Attention Learning

Multi-Label Chest X-Ray Image Classification via Semantic Similarity Graph Embedding

Modeling Global and Local Label Correlation with Graph Convolutional Networks for Multi-Label Chest X-ray Image Classification

Quantifying and Leveraging Classification Uncertainty for Chest Radiograph Assessment

Chest X-rays Classification: A Multi-Label and Fine-Grained Problem

Cross Chest Graph for Disease Diagnosis with Structural Relational Reasoning

Deep Learning Based Classification of Multi-Label Chest X-ray Images Via Dual-Weighted Metric Loss

Chest X-ray Image Classification: A Causal Perspective

MRChexNet: Multi-modal bridge and relational learning for thoracic disease recognition in chest X-rays

Multilabel Classification for Lung Disease Detection: Integrating Deep Learning and Natural Language Processing

Dynamic Graph Correlation Learning for Disease Diagnosis with Incomplete Labels

MS-ANet: Deep Learning for Automated Multi-Label Thoracic Disease Detection and Classification.

Multi-Label Generalized Zero Shot Chest Xray Classification By Combining Image-Text Information With Feature Disentanglement

Multi-Label Local to Global Learning: A Novel Learning Paradigm for Chest X-Ray Abnormality Classification.

Label Semantic Improvement with Graph Convolutional Networks for Multi-Label Chest X-Ray Image Classification

Triple Attention Learning for Classification of 14 Thoracic Diseases Using Chest Radiography.

Learning To Recognize Thoracic Disease In Chest X-Rays With Knowledge-Guided Deep Zoom Neural Networks

Towards long-tailed, multi-label disease classification from chest X-ray: Overview of the CXR-LT challenge