Deep GRU-CNN Model for COVID-19 Detection From Chest X-Rays Data

Pir Masoom Shah, Faizan Ullah, Dilawar Shah, Abdullah Gani, Carsten Maple, Yulin Wang, Shahid, Muhammad Abrar, Saif ul Islam

TL;DR

This paper introduces a novel deep learning model, a hybrid GRU-CNN architecture, for detecting COVID-19 from chest X-ray (CXR) images. By combining the feature extraction capabilities of CNNs with the sequential data processing power of GRUs, the model achieves high accuracy in classifying COVID-19, pneumonia, and normal cases, offering a promising tool for rapid and automated diagnosis.

What Problem Does This Paper Address?

The COVID-19 pandemic has highlighted the critical need for rapid and accurate diagnostic tools. Traditional methods like RT-PCR, while effective, can be time-consuming and resource-intensive, especially during widespread outbreaks. This strain on healthcare systems necessitates automated solutions for early detection. Chest X-rays (CXRs) are a readily available imaging modality, but subtle signs of COVID-19 can be missed by human interpretation, leading to potential misdiagnosis and delayed treatment. This paper aims to address this challenge by developing a deep learning model that can accurately and efficiently detect COVID-19 from CXR images, thereby assisting medical practitioners in timely diagnosis and potentially mitigating the spread of the virus.

Methods

The core of this research is a hybrid deep learning model that combines a Convolutional Neural Network (CNN) with a Gated Recurrent Unit (GRU).

  • CNN for Feature Extraction: The CNN component is responsible for learning and extracting relevant features from the CXR images. The paper details a CNN architecture with ten convolutional layers and five max-pooling layers. This hierarchical structure allows the network to progressively learn complex patterns, from simple edges to more intricate textures indicative of lung abnormalities.
  • GRU for Classification: Following feature extraction by the CNN, the extracted features are fed into a GRU. GRUs are a type of Recurrent Neural Network (RNN) particularly adept at handling sequential data and mitigating the vanishing gradient problem common in traditional RNNs. In this context, the GRU acts as a classifier, leveraging the sequential nature of the extracted features to make a final prediction about the presence of COVID-19, pneumonia, or a normal condition. The authors specifically chose GRU to address potential issues with limited data, where CNNs can sometimes struggle with weight vanishing.
  • Attention Maps (Grad-CAM): To enhance the interpretability of the model, the researchers employed Gradient-weighted Class Activation Mapping (Grad-CAM). This technique generates visual heatmaps that highlight the regions in the CXR image that the model focused on when making its prediction, providing insights into why the model arrived at a particular diagnosis and aiding human experts in verification.

Datasets

The study utilized a dataset of 424 chest X-ray (CXR) images. These images were sourced from two different repositories:

  • Joseph Paul Cohen's repository: This source provided COVID-19 infected X-ray images.
  • Kaggle repository: This source provided images for pneumonia and normal cases.

To ensure class balance, the dataset was curated to have an equal number of instances (141) for each of the three classes: COVID-19, Normal, and Pneumonia. The dataset was then split into training (70%), validation (10%), and testing (20%) sets.

Key Insights and Ideas

  • Hybrid CNN-GRU Synergy: The paper demonstrates the effectiveness of combining CNNs for spatial feature extraction with GRUs for sequential classification. This hybrid approach leverages the strengths of both architectures to achieve robust performance on medical image classification tasks.
  • Addressing Data Scarcity: The authors explicitly mention the challenge of limited publicly available data for COVID-19 research. Their choice of GRU was partly motivated by its ability to handle such constraints, and the use of Grad-CAM adds a layer of interpretability that can be crucial when dealing with smaller datasets.
  • Interpretability for Medical AI: The integration of Grad-CAM is a significant step towards making AI models more trustworthy in clinical settings. By visualizing the model's decision-making process, it allows medical professionals to understand and validate the AI's output, fostering greater confidence in its use.
  • Early Detection as a Public Health Imperative: The paper strongly emphasizes the critical role of early detection in controlling the spread of infectious diseases like COVID-19, highlighting the societal impact of their research.

Key Results

The proposed Deep GRU-CNN model demonstrated strong performance on the test set:

  • Overall Performance: The model achieved mean precision, recall, and F1-score of 0.96, 0.96, and 0.95, respectively, across all three classes.
  • Class-wise Performance:
    • Normal: Precision: 0.90, Recall: 1.00, F1-score: 0.95
    • COVID-19: Precision: 1.00, Recall: 1.00, F1-score: 1.00
    • Pneumonia: Precision: 1.00, Recall: 0.89, F1-score: 0.94
  • ROC Curve Analysis: The model achieved high ROC scores for all classes, with COVID-19 scoring 1.00, Normal scoring 0.9969, and Pneumonia scoring 0.9962, indicating excellent discriminative capabilities.
  • Confusion Matrix: The confusion matrix showed that the model correctly classified most instances, with only a few misclassifications, particularly in the pneumonia class where 3 out of 28 images were misclassified as normal.

Limitations and Future Work

The authors acknowledge several limitations:

  • Limited Views: The model is currently limited to analyzing X-ray images from the posterior-anterior (PA) view only. Other views (e.g., AP, lateral) and factors are not evaluated.
  • Data Availability: The model was trained and tested on a relatively small dataset due to the limited availability of COVID-19 data.
  • Human Expert Comparison: The performance comparison was made with previous algorithms, but a direct comparison with human domain experts (radiologists) was not conducted.

For future work, the authors intend to:

  • Generative Adversarial Models: Explore the use of Generative Adversarial Models (GANs) for data augmentation to further improve model performance, especially given the data scarcity.
  • Expand to Other Views: Extend the model's capability to analyze other X-ray views and potentially incorporate other imaging modalities.

Summary updated

Abstract

In the current era, data is growing exponentially due to advancements in smart devices. Data scientists apply a variety of learning-based techniques to identify underlying patterns in the medical data to address various health-related issues. In this context, automated disease detection has now become a central concern in medical science. Such approaches can reduce the mortality rate through accurate and timely diagnosis. COVID-19 is a modern virus that has spread all over the world and is affecting millions of people. Many countries are facing a shortage of testing kits, vaccines, and other resources due to significant and rapid growth in cases. In order to accelerate the testing process, scientists around the world have sought to create novel methods for the detection of the virus. In this paper, we propose a hybrid deep learning model based on a convolutional neural network (CNN) and gated recurrent unit (GRU) to detect the viral disease from chest X-rays (CXRs). In the proposed model, a CNN is used to extract features, and a GRU is used as a classifier. The model has been trained on 424 CXR images with 3 classes (COVID-19, Pneumonia, and Normal). The proposed model achieves encouraging results of 0.96, 0.96, and 0.95 in terms of precision, recall, and f1-score, respectively. These findings indicate how deep learning can significantly contribute to the early detection of COVID-19 in patients through the analysis of X-ray scans. Such indications can pave the way to mitigate the impact of the disease. We believe that this model can be an effective tool for medical practitioners for early diagnosis.
BibTeX
@article{shah2022deep,
  author = {Pir Masoom Shah and Faizan Ullah and Dilawar Shah and Abdullah Gani and Carsten Maple and Yulin Wang and Shahid and Muhammad Abrar and Saif ul Islam},
  title = {{Deep GRU-CNN Model for COVID-19 Detection From Chest X-Rays Data}},
  journal = {IEEE Access},
  year = {2022},
  doi = {10.1109/access.2021.3077592},
}
Figure