What is it about?
The outbreak of the SARS-CoV-2/Covid-19 virus in 2019-2020 has made the world look for fast and accurate detection methods of the disease. The most commonly used tools for detecting Covid patients are Chest-X-ray or Chest-CT-scans of the patient. However, sometimes it’s hard for the physicians to diagnose the SARS-CoV-2 infection from the raw image. Moreover, sometimes, deep-learning-based techniques, using raw images, fail to detect the infection. Hence, this paper represents a hybrid method employing both traditional signal processing and deep learning technique for quick detection of SARS-CoV-2 patients based on the CT-scan and Chest-X-ray images of a patient.
Featured Image
Photo by Fusion Medical Animation on Unsplash
Why is it important?
In all of the deep-learning-based studies, either the raw input image or any lightly preprocessed image has been used as input data to the deep learning network. However, we wanted to explore this preprocessing task, where the raw image is processed by the mode decomposition technique so that it is easier for the network to learn the inherent features in a more effective way. In this study, two-dimensional Empirical Mode Decomposition (2DEMD) has been presented as a new strategy to extract features from CT-scan and Chest-X-ray images. 2DEMD has never been used as a preprocessing technique for deep-learning-based methods. The primary goal of the classification is to show that a modified image from 2DEMD performs better as input data for the deep CNN while classifying Covid patients. The paper represents that training a fundamental deep neural network with a modified CT-scan/Chest-X-ray (through 2DEMD) shows better performance than training with a raw image. The modified image acts as a better candidate than the raw image irrespective of the complexity and performance of the CNN and the variation of data in the databases. The proposed method has been validated on three publicly available SARS-CoV-2 databases - two comprised of CT-scan images, and the third consists of Chest-X-ray images.
Perspectives
I hope this article helps people look into the covid research from a different perspective. Although this is a small contribution to a massive problem, it might help data scientists and doctors visualize the covid patients' lung condition from a new paradigm.
Nahian Ibn Hasan
Purdue University
Read the Original
This page is a summary of: A Hybrid Method of Covid-19 Patient Detection From Modified CT-Scan/Chest-X-Ray Images Combining Deep Convolutional Neural Network And Two- Dimensional Empirical Mode Decomposition, Computer Methods and Programs in Biomedicine Update, July 2021, Elsevier,
DOI: 10.1016/j.cmpbup.2021.100022.
You can read the full text:
Resources
Open Access Article
The outbreak of the SARS-CoV-2/Covid-19 virus in 2019-2020 has made the world look for fast and accurate detection methods of the disease. The most commonly used tools for detecting Covid patients are Chest-X-ray or Chest-CT-scans of the patient. However, sometimes it’s hard for the physicians to diagnose the SARS-CoV-2 infection from the raw image. Moreover, sometimes, deep-learning-based techniques, using raw images, fail to detect the infection. Hence, this paper represents a hybrid method employing both traditional signal processing and deep learning technique for quick detection of SARS-CoV-2 patients based on the CT-scan and Chest-X-ray images of a patient. Unlike the other AI-based methods, here, a CT-scan/Chest-X-ray image is decomposed by two-dimensional Empirical Mode Decomposition (2DEMD), and it generates different orders of Intrinsic Mode Functions (IMFs). Next, The decomposed IMF signals are fed into a deep Convolutional Neural Network (CNN) for feature extraction and classification of Covid patients and Non-Covid patients. The proposed method is validated on three publicly available SARS-CoV-2 data sets using two deep CNN architectures. In all the databases, the modified CT-scan/Chest-X-ray image provides a better result than the raw image in terms of classification accuracy of two fundamental CNNs. This paper represents a new viewpoint of extracting preprocessed features from the raw image using 2DEMD.
Graphical Abstract
The proposed method for SARS-CoV-2 patient classification.
Contributors
The following have contributed to this page
