[Paper Review] Robust Screening of COVID-19 from Chest X-ray via Discriminative Cost-Sensitive Learning

Tianyang Li, Zhongyi Han+4 more|arXiv (Cornell University)|Apr 27, 2020

COVID-19 diagnosis using AIMedicine56 references48 citations

TL;DR

The paper introduces discriminative cost-sensitive learning (DCSL) for COVID-19 screening from chest X-rays, combining a conditional center loss and score-level cost-sensitive learning to improve fine-grained classification and reduce misdiagnosis costs, achieving 97.01% accuracy on a three-class dataset.

ABSTRACT

This paper addresses the new problem of automated screening of coronavirus disease 2019 (COVID-19) based on chest X-rays, which is urgently demanded toward fast stopping the pandemic. However, robust and accurate screening of COVID-19 from chest X-rays is still a globally recognized challenge because of two bottlenecks: 1) imaging features of COVID-19 share some similarities with other pneumonia on chest X-rays, and 2) the misdiagnosis rate of COVID-19 is very high, and the misdiagnosis cost is expensive. While a few pioneering works have made much progress, they underestimate both crucial bottlenecks. In this paper, we report our solution, discriminative cost-sensitive learning (DCSL), which should be the choice if the clinical needs the assisted screening of COVID-19 from chest X-rays. DCSL combines both advantages from fine-grained classification and cost-sensitive learning. Firstly, DCSL develops a conditional center loss that learns deep discriminative representation. Secondly, DCSL establishes score-level cost-sensitive learning that can adaptively enlarge the cost of misclassifying COVID-19 examples into other classes. DCSL is so flexible that it can apply in any deep neural network. We collected a large-scale multi-class dataset comprised of 2,239 chest X-ray examples: 239 examples from confirmed COVID-19 cases, 1,000 examples with confirmed bacterial or viral pneumonia cases, and 1,000 examples of healthy people. Extensive experiments on the three-class classification show that our algorithm remarkably outperforms state-of-the-art algorithms. It achieves an accuracy of 97.01%, a precision of 97%, a sensitivity of 97.09%, and an F1-score of 96.98%. These results endow our algorithm as an efficient tool for the fast large-scale screening of COVID-19.

Motivation & Objective

• Address the challenges of distinguishing COVID-19 from other pneumonia and healthy X-rays in a three-class setting.
• Develop a discriminative, cost-sensitive framework that emphasizes COVID-19 risk to reduce misdiagnosis costs.
• Leverage fine-grained representation learning and class-aware costs to improve screening performance.

Proposed method

• Introduce conditional center loss to learn class-balanced, discriminative representations for fine-grained COVID-19 classification.
• Propose score-level cost-sensitive learning by applying a domain-informed cost matrix after the model outputs to bias predictions toward COVID-19 when appropriate.
• Combine the two modules into the Discriminative Cost-Sensitive Learning (DCSL) framework with end-to-end optimization.
• Evaluate on a multi-center chest X-ray dataset with 239 COVID-19, 1,000 bacterial/viral pneumonia, and 1,000 healthy images.
• Use VGG16 as backbone, transfer learning from ImageNet, data augmentation, and 5-fold cross-validation.

Experimental results

Research questions

• RQ1 Can a fine-grained, cost-sensitive approach improve COVID-19 screening from chest X-rays over standard loss functions?
• RQ2 Does incorporating class-conditional information into center loss and a domain-informed score-level cost matrix reduce misclassification of COVID-19?
• RQ3 How does DCSL perform in comparison to established CNN architectures and COVID-specific models on a three-class chest X-ray task?

Key findings

• DCSL achieves 97.01% accuracy, 97.00% precision, 97.09% sensitivity, and 96.98% F1-score on the three-class task.
• DCSL outperforms COVID-Net and other baselines across all reported metrics.
• Ablation studies show that conditional center loss and score-level cost-sensitive learning contribute substantially to performance gains.
• Three-class setup (COVID-19, healthy, other pneumonia) demonstrates strong discriminability with the proposed approach.