[Paper Review] Comparative Analysis of State-of-the-Art Deep Learning Models for Detecting COVID-19 Lung Infection from Chest X-Ray Images

Zeba Ghaffar, Pir Masoom Shah+6 more|arXiv (Cornell University)|Jul 1, 2022

COVID-19 diagnosis using AI4 citations

TL;DR

This study evaluates the performance of ten state-of-the-art deep convolutional neural networks (CNNs) for classifying COVID-19 from chest X-ray images, achieving up to 95% average accuracy using MobileNet and EfficientNet. The models are explainable via Grad-CAM visualization, enhancing clinical trust and interpretability in automated diagnosis.

ABSTRACT

The ongoing COVID-19 pandemic has already taken millions of lives and damaged economies across the globe. Most COVID-19 deaths and economic losses are reported from densely crowded cities. It is comprehensible that the effective control and prevention of epidemic/pandemic infectious diseases is vital. According to WHO, testing and diagnosis is the best strategy to control pandemics. Scientists worldwide are attempting to develop various innovative and cost-efficient methods to speed up the testing process. This paper comprehensively evaluates the applicability of the recent top ten state-of-the-art Deep Convolutional Neural Networks (CNNs) for automatically detecting COVID-19 infection using chest X-ray images. Moreover, it provides a comparative analysis of these models in terms of accuracy. This study identifies the effective methodologies to control and prevent infectious respiratory diseases. Our trained models have demonstrated outstanding results in classifying the COVID-19 infected chest x-rays. In particular, our trained models MobileNet, EfficentNet, and InceptionV3 achieved a classification average accuracy of 95\%, 95\%, and 94\% test set for COVID-19 class classification, respectively. Thus, it can be beneficial for clinical practitioners and radiologists to speed up the testing, detection, and follow-up of COVID-19 cases.

Motivation & Objective

• To evaluate the effectiveness of the top ten state-of-the-art deep CNNs in detecting COVID-19 from chest X-ray images.
• To compare model performance in terms of classification accuracy for COVID-19, normal, and pneumonia cases.
• To enhance model interpretability through Grad-CAM visualization for clinical validation and trust.
• To provide a practical, fast, and cost-effective alternative to traditional RT-PCR testing for early COVID-19 detection.
• To support radiologists and clinicians with AI-assisted tools for faster diagnosis and patient follow-up.

Proposed method

• Trained ten state-of-the-art deep CNN architectures, including MobileNet, EfficientNet, InceptionV3, DenseNet, ResNet, and NASNet variants, on a chest X-ray dataset.
• Employed transfer learning with ImageNet-pretrained weights to fine-tune models for the binary (COVID-19 vs. normal/pneumonia) and multi-class classification tasks.
• Utilized data augmentation techniques to improve generalization and reduce overfitting during training.
• Applied Grad-CAM (Gradient Class Activation Mapping) to generate attention maps visualizing the regions of the X-ray that influenced the model’s predictions.
• Evaluated model performance using standard metrics including accuracy, precision, recall, and F1-score on a held-out test set.
• Conducted qualitative analysis by visualizing Grad-CAM maps for both correctly and misclassified cases to assess model reasoning and reliability.

Experimental results

Research questions

• RQ1 Which state-of-the-art deep learning models achieve the highest classification accuracy in detecting COVID-19 from chest X-ray images?
• RQ2 How do different CNN architectures perform across multiple classes (COVID-19, normal, pneumonia) in terms of precision and recall?
• RQ3 To what extent can Grad-CAM explainability improve trust in AI-based diagnosis among medical professionals?
• RQ4 Can deep learning models detect subtle pathological patterns in chest X-rays that correlate with clinical expertise, particularly in lung regions?
• RQ5 What are the failure modes of these models, and how do attention maps reveal misclassification patterns?

Key findings

• MobileNet and EfficientNet achieved the highest average test accuracy of 95% for classifying COVID-19 from chest X-rays, outperforming other models.
• InceptionV3 achieved a 94% average test accuracy, demonstrating strong performance despite its deeper architecture.
• Grad-CAM visualizations confirmed that models focus on lung regions—consistent with radiological expertise—when classifying COVID-19 cases.
• For normal X-rays, models primarily attended to rib structures, aligning with expected anatomical features.
• In pneumonia cases, the models highlighted upper lobes of the lungs, matching known imaging patterns of pulmonary infection.
• Misclassified examples revealed that attention maps sometimes focused on non-pulmonary or non-specific regions, indicating potential limitations in generalization under ambiguous cases.