Performance Evaluation of Deep Learning-based Covid-19 Diagnosis Software: a Comprehensive Approach Using Convolutional Neural Networks and Ensemble Machine Learning

Abstract

Abstract Quick diagnosis of COVID-19 through chest X-ray images has gained significant attention due to its potential to aid in rapid screening. In this study, we presented a comprehensive approach utilizing convolutional neural networks (CNNs) for feature extraction from chest X-ray images, followed by an ensemble of classifiers including Decision Tree, Support Vector Machine, Random Forest, and AdaBoost for accurate classification. Our CNN architecture, trained on Google Colab with GPU runtime, comprises 20 layers incorporating Conv2D, MaxPooling2D, Dropout, and fully connected layers with ReLU activation function and a dropout threshold of 0.25, achieving an accuracy of 97.10%. By using a dataset that consists of 33,920 chest X-ray (CXR) images including 11,956 COVID-19, 11,263 Non-COVID infections (Viral or Bacterial Pneumonia), 10,701 Normal with Ground-truth lung segmentation masks provided for the entire dataset from the Kaggle COVID-19 Radiography Database. Our final ensemble classifier, employing Soft voting, attained a heightened accuracy of 97.51%. Moreover, to gain insights into the CNN's internal processes, we visualized intermediate layer activations. Subsequently, we deployed the final model using a Flask API for seamless integration into healthcare systems. Our approach promised efficient and accurate diagnosis of COVID-19 from chest X-ray images, facilitating timely patient management.