COVID-19 Diagnosis by Stationary Wavelet Entropy and Extreme Learning Machine

Introduction

An outbreak of coronavirus disease 19 (COVID-19) infection began in December 2019. According to statistics released by Johns Hopkins University in the United States, as of July 15, Beijing time, the number of confirmed cases of new coronary pneumonia worldwide rose to 560,541,147, of which 6,366,285 died. The most common clinical symptoms are fever and cough, in addition to other nonspecific symptomatology including dyspnea, headache, muscle soreness, and fatigue (Wang et al., 2020).

Chest computed tomography (CT) (Kulkarni et al., 2022) is an effective tool for diagnosing COVID-19. It is faster and more sensitive (Ai, 2020). The main biomarkers differentiating COVID-19 patients from healthy individuals are asymmetric peripheral ground-glass opacities (GGOs) without pleural effusion (Li & Xia, 2020). Doctors manually interpret and diagnose COVID-19 based on CT images each time. The workload is large, and there are many external factors, which are prone to misdiagnosis. This paper designs an intelligent diagnostic system that combines computer vision and artificial intelligence. It provides convenience for patients, doctors and hospitals.

Similar intelligent systems have achieved excellent results in analyzing images. Lu (2016) designed a radial basis function neural network (RBFNN) to detect pathological brains. Lu, S. (2018) proposed kernel based extreme learning machine (K-ELM) to design an intelligent system for detecting pathological brains. K-ELM can provide different kernel functions for different applications such as optimization, regression and classification. Lu (2017) proposed an extreme learning machine using the bat algorithm (ELM-BA) for training datasets. It solves the problems of redundant nodes and non-optimal weights/biases. Two papers focused on evaluating machine learning techniques for diagnosing COVID-19 [8, 9]. Yao (2020) built an artificial intelligence model to analyze CT scan images. It is based on wavelet entropy (WE) and biogeography-based optimization (BBO). It can accurately detect COVID-19. Chen (2020) proposed a system based on Gray-Level Co-Occurrence Matrix (GLCM) and Support Vector Machine (SVM). It can classify chest CT images efficiently and accurately to identify COVID-19. Wang (2021) builds a machine learning model based on wavelet entropy (WE) and Jaya algorithm. It can quickly and accurately analyze CT images to identify COVID-19 patients. Wang (2022) proposes a classification method of COVID-19 chest CT images. This is an algorithm based on wavelet entropy (WE) and cat swarm optimization (CSO).

Based on a dataset of chest CT images, this study proposes a new method for diagnosing COVID-19. Image features were extracted by stationary wavelet entropy (SWE), the data were classified and trained using extreme learning machine algorithm (ELM), and an appropriate model was determined by k-fold cross-validation (k-fold CV). The goal is to obtain more accurate chest CT images of COVID-19 patients without redundancy (Zhang et al., 2021), helping clinicians to make more precise judgments. Compared with the current state-of-the-art approaches, the experimental results show a more accurate advantage.

The contributions of this paper are:

• 1.

  Introduce the method of extracting CT image features by SWE.

• 2.

  Introduces the method for classifying input image datasets by ELM.

• 3.

  Present a more precise method for diagnosing COVID-19 by SWE and ELM based on chest CT images.