Source-linked AI summary
COVIDX-Net: A Framework of Deep Learning Classifiers to Diagnose COVID-19 in X-Ray Images
Ezz El-Din Hemdan, Marwa A. Shouman, Mohamed Esmail Karar
TL;DR
The paper addresses limited evidence for identifying COVID-19 in X-ray images and the need for diagnostic support alongside constrained imaging options. It proposes COVIDX-Net, a framework combining deep-learning classifiers for automated classification, and reports that VGG19 and DenseNet201 achieved the strongest performance scores. The authors position the framework as a basis for further development, with clinical studies identified as the next milestone.
Problem
Application of deep-learning techniques to identify novel COVID-19 in X-ray images was still very limited, while CT involved high patient dose and scan cost.
Method
COVIDX-Net builds a framework of deep-learning models to automatically classify COVID-19 status in 2-D chest X-ray images.
Results
VGG19 and DenseNet201 achieved the best performance scores among the tested deep-learning classifiers.
Takeaways & Limitations
The study demonstrates a useful application of deep-learning models for classifying COVID-19 in X-ray images and identifies clinical studies as the next milestone.
Takeaways & Limitations
The introduction identifies CT's high patient dose and scan cost as a boundary motivating the X-ray-focused approach.
Abstract
from arXiv · showhide
Background and Purpose: Coronaviruses (CoV) are perilous viruses that may cause Severe Acute Respiratory Syndrome (SARS-CoV), Middle East Respiratory Syndrome (MERS-CoV). The novel 2019 Coronavirus disease (COVID-19) was discovered as a novel disease pneumonia in the city of Wuhan, China at the end of 2019. Now, it becomes a Coronavirus outbreak around the world, the number of infected people and deaths are increasing rapidly every day according to the updated reports of the World Health Organization (WHO). Therefore, the aim of this article is to introduce a new deep learning framework; namely COVIDX-Net to assist radiologists to automatically diagnose COVID-19 in X-ray images. Materials and Methods: Due to the lack of public COVID-19 datasets, the study is validated on 50 Chest X-ray images with 25 confirmed positive COVID-19 cases. The COVIDX-Net includes seven different architectures of deep convolutional neural network models, such as modified Visual Geometry Group Network (VGG19) and the second version of Google MobileNet. Each deep neural network model is able to analyze the normalized intensities of the X-ray image to classify the patient status either negative or positive COVID-19 case. Results: Experiments and evaluation of the COVIDX-Net have been successfully done based on 80-20% of X-ray images for the model training and testing phases, respectively. The VGG19 and Dense Convolutional Network (DenseNet) models showed a good and similar performance of automated COVID-19 classification with f1-scores of 0.89 and 0.91 for normal and COVID-19, respectively. Conclusions: This study demonstrated the useful application of deep learning models to classify COVID-19 in X-ray images based on the proposed COVIDX-Net framework. Clinical studies are the next milestone of this research work.
1. Introduction
COVID-19 created a need for timely diagnostic support, while CT imaging carries higher dose and cost and X-ray has limited soft-tissue contrast. The study therefore proposes COVIDX-Net, a framework of deep-learning classifiers to assist radiologists in identifying COVID-19 from X-ray images.
- COVID-19 emerged as a novel pneumonia and spread internationally, creating an urgent diagnostic context.
- RT-PCR was described as time-consuming and associated with high false-negative levels, motivating complementary imaging approaches.
- CT can support COVID-19 confirmation but is constrained by high patient dose and scan cost, whereas chest X-ray machines are broadly available.
- Chest X-ray is commonly the first-choice modality for radiologists, although limited soft-tissue contrast restricts its exposure dose.
- Deep convolutional neural networks have been applied to medical-image classification, but their use for detecting novel COVID-19 in X-ray images remained limited.
- COVIDX-Net combines pretrained deep-learning classifiers to assist automatic COVID-19 diagnosis in X-ray images and compare model performance on a small dataset.
2. Methods
COVIDX-Net combines seven deep convolutional architectures in a workflow that preprocesses chest X-rays, trains and validates classifiers, and evaluates binary COVID-19 classification.
- Proposed COVIDX-Net Description: Seven architectures—VGG19, DenseNet201, InceptionV3, ResNetV2, InceptionResNetV2, Xception, and MobileNetV2—form the COVIDX-Net classifier framework.The framework uses these alternative deep convolutional neural network designs for conventional X-ray images.
- Proposed COVIDX-Net Description: X-ray images are resized to 224 X 224 pixels and labels are one-hot encoded for positive COVID-19 or “not” cases.These preprocessing steps prepare images and labels for the deep learning pipeline.
- Training Model and Validation: The dataset is split into 80% training and validation data and 20% testing data, with subsampling used during classifier training.The 80% portion is divided into equal training and validation sets.
- Training Model and Validation: Tuned classifiers categorize test images as confirmed positive COVID-19 or normal, after which each model’s performance is evaluated.The workflow feeds testing data to the tuned classifier and assesses results using defined metrics.
- Classification Performance Analysis: Performance analysis uses confusion-matrix outcomes and accuracy, precision, recall, and F1-score to assess classification quality.The study describes true and false outcomes and applies these metrics to evaluate the classifiers.
- Classification Performance Analysis: Accuracy is calculated from true positives and true negatives divided by all confusion-matrix components, while F1-score combines precision and recall.The authors also emphasize considering false-positive and false-negative values when assessing reliability.
3 Experiments
The experiments evaluated seven deep learning classifiers on a small, balanced X-ray dataset using an 80–20% training-testing split. VGG19 and DenseNet201 achieved the strongest accuracy, while MobileNetV2 offered the fastest computation.
- 3.1 Dataset and Experimental setup: The dataset contained 50 X-ray images: 25 normal cases and 25 positive COVID-19 cases.Images originally ranged from 1112 × 624 to 2170 × 1953 pixels and were resized to 224 × 224 pixels.
- 3.1 Dataset and Experimental setup: The experiments randomly assigned 40 images to training and 10 to testing, corresponding to an 80–20% split.The models were trained with SGD, a learning rate of e−3, batch size 7, and 50 epochs; data augmentation was not used.
- 3.2 Overall Performance Evaluation: 389.0 and 1.0 seconds were the MobileNetV2 training and testing times, respectively, making it the fastest tested model despite 60% accuracy.Across all models, reported running times ranged from 390.0 to 2645.0 seconds, and testing on 10 images did not exceed 6 seconds in the cited evaluation.
- 3.2 Overall Performance Evaluation: ResNetV2, InceptionResNetV2, Xception, and MobileNetV2 achieved the highest precision for positive COVID-19 detection but performed worst on normal-case classification.The authors therefore recommended VGG19 and DenseNet201 for CAD systems identifying patient health status from X-ray images.
4 Conclusions
The conclusion presents COVIDX-Net as a framework using seven deep learning classifiers to identify COVID-19 in 2-D X-ray images. VGG19 and DenseNet201 produced the best performance scores, while MobileNetV2 was noted for high computational speed and possible smart-device use.
- 4 Conclusions: VGG19 and DenseNet201 produced the best performance scores among the evaluated classifiers.The conclusion identifies these models as the strongest-performing models in the proposed framework.
- 4 Conclusions: MobileNetV2 was highlighted for its high computational speed and potential use on smart devices in healthcare.The authors state that its performance could be further improved for this application.
- 4 Conclusions: The authors identify CT imaging as a possible topic for future work.This is presented alongside the framework’s current focus on X-ray imaging.