[Paper Review] Towards an Efficient Deep Learning Model for COVID-19 Patterns Detection in X-ray Images

Eduardo Luz, Pedro Silva+2 more|arXiv (Cornell University)|Apr 12, 2020

COVID-19 diagnosis using AIMedicine37 citations

TL;DR

This paper proposes an efficient deep learning model based on the EfficientNet architecture with a hierarchical classifier to detect COVID-19 in chest X-rays. It achieves 93.9% overall accuracy, 96.8% sensitivity, and 100% positive predictive value for COVID-19 while using 5 to 30 times fewer parameters than competing models, significantly reducing computational cost.

ABSTRACT

Confronting the pandemic of COVID-19, is nowadays one of the most prominent challenges of the human species. A key factor in slowing down the virus propagation is the rapid diagnosis and isolation of infected patients. The standard method for COVID-19 identification, the Reverse transcription polymerase chain reaction method, is time-consuming and in short supply due to the pandemic. Thus, researchers have been looking for alternative screening methods and deep learning applied to chest X-rays of patients has been showing promising results. Despite their success, the computational cost of these methods remains high, which imposes difficulties to their accessibility and availability. Thus, the main goal of this work is to propose an accurate yet efficient method in terms of memory and processing time for the problem of COVID-19 screening in chest X-rays. Methods: To achieve the defined objective we exploit and extend the EfficientNet family of deep artificial neural networks which are known for their high accuracy and low footprints in other applications. We also exploit the underlying taxonomy of the problem with a hierarchical classifier. A dataset of 13,569 X-ray images divided into healthy, non-COVID-19 pneumonia, and COVID-19 patients is used to train the proposed approaches and other 5 competing architectures. Finally, 231 images of the three classes were used to assess the quality of the methods. Results: The results show that the proposed approach was able to produce a high-quality model, with an overall accuracy of 93.9%, COVID-19, sensitivity of 96.8% and positive prediction of 100%, while having from 5 to 30 times fewer parameters than other than the other tested architectures. Larger and more heterogeneous databases are still needed for validation before claiming that deep learning can assist physicians in the task of detecting COVID-19 in X-ray images.

Motivation & Objective

• Address the urgent need for rapid and accessible COVID-19 screening methods due to limitations of RT-PCR testing.
• Reduce the high computational cost of existing deep learning models for X-ray-based COVID-19 detection to improve accessibility.
• Develop a model that maintains high accuracy while minimizing memory and processing time requirements.
• Validate the model on a diverse dataset of 13,569 X-ray images across three classes: healthy, non-COVID-19 pneumonia, and COVID-19.
• Explore the potential of hierarchical classification to improve model efficiency and performance in multi-class medical image analysis.

Proposed method

• Adapt and extend the EfficientNet family of deep neural networks, known for high accuracy and low computational footprint, for COVID-19 detection in X-ray images.
• Implement a hierarchical classifier to structure the classification task into multiple levels, improving efficiency and decision clarity.
• Train the model on a dataset of 13,569 X-ray images labeled as healthy, non-COVID-19 pneumonia, or COVID-19.
• Evaluate performance on a separate test set of 231 images across the three classes to ensure robustness and generalization.
• Compare the proposed model’s performance and parameter count against five other deep learning architectures.
• Leverage transfer learning principles by initializing with pre-trained ImageNet weights to accelerate convergence and improve performance.

Experimental results

Research questions

• RQ1 Can an EfficientNet-based model achieve high diagnostic accuracy for COVID-19 detection in chest X-rays while maintaining low computational cost?
• RQ2 How does hierarchical classification improve the efficiency and performance of multi-class X-ray analysis for COVID-19?
• RQ3 To what extent does the proposed model reduce parameter count compared to existing deep learning architectures without sacrificing accuracy?
• RQ4 What is the sensitivity and positive predictive value of the model in detecting actual COVID-19 cases in a real-world test set?
• RQ5 How generalizable is the model given current dataset limitations, and what further data is needed for broader clinical validation?

Key findings

• The proposed model achieved an overall accuracy of 93.9% on the test set of 231 X-ray images.
• The model demonstrated a high sensitivity of 96.8% for detecting COVID-19 cases, indicating strong ability to identify true positives.
• The positive predictive value reached 100%, meaning all predicted COVID-19 cases were correctly identified in the test set.
• The model used between 5 and 30 times fewer parameters than the other five tested deep learning architectures, significantly reducing memory and processing demands.
• Despite strong performance, the study acknowledges the need for larger and more diverse datasets to support broader clinical deployment.
• The combination of EfficientNet and hierarchical classification proved effective in balancing accuracy and efficiency for medical imaging applications.