Detection of COVID19 from X-ray images using multiscale Deep Convolutional Neural Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• The COVID19 pandemic poses a significant global health challenge, necessitating rapid and accurate diagnostic tools.
• Traditional COVID19 diagnostic methods like RT-PCR are time-consuming and costly, while manual interpretation of X-rays is labor-intensive.
• Computer-aided diagnosis (CAD) systems offer a promising solution for automated and efficient disease detection.

Purpose of the Study:

• To develop and evaluate a novel automated system for detecting COVID19 using chest X-ray images.
• To improve the speed and accuracy of COVID19 diagnosis compared to existing methods.
• To classify X-ray images into no-finding, pneumonia, and COVID19 categories.

Main Methods:

• A novel approach utilizing Fixed Boundary-based Two-Dimensional Empirical Wavelet Transform (FB2DEWT) to extract modes from X-ray images.
• Decomposition of each X-ray image into seven modes using FB2DEWT.
• Inputting the extracted modes into a multiscale deep Convolutional Neural Network (CNN) for classification.

Main Results:

• The proposed deep learning model achieved a maximum accuracy of 96% for multiclass and 100% for binary classification on dataset A (1225 images) with 5-fold cross-validation.
• On dataset B (9000 images), accuracies of 97.17% (multiclass) and 96.06% (binary) were obtained using the multiscale deep CNN with 5-fold cross-validation.
• The system demonstrated superior classification performance compared to existing approaches for COVID19 detection from X-ray images.

Conclusions:

• The proposed automated system effectively detects COVID19 using chest X-ray images.
• The combination of FB2DEWT and multiscale deep CNN offers a highly accurate and efficient method for COVID19 diagnosis.
• This approach has the potential to aid clinicians in rapid and reliable COVID19 detection, especially when laboratory tests are limited.