Here, we use multi-type feature fusion and selection to predict COVID-19 infections on chest computed tomography (CT) scans. The scheme operates in four steps. Initially, we prepared a database containing COVID-19 pneumonia and normal CT scans. These images were retrieved from the Radiopaedia COVID-19 website. The images were divided into training and test sets in a ratio of 70:30. Then, multiple features were extracted from the training data. We used canonical correlation analysis to fuse the features into single vectors; this enhanced the predictive capacity. We next implemented a genetic algorithm (GA) in which an Extreme Learning Machine (ELM) served to assess GA fitness. Based on the ELM losses, the most discriminatory features were selected and saved as an ELM Model. Test images were sent to the model, and the best-selected features compared to those of the trained model to allow final predictions. Validation employed the collected chest CT scans. The best predictive accuracy of the ELM classifier was 93.9%; the scheme was effective.
The novel coronavirus pandemic disease that appeared in China has rapidly spread worldwide [
To date (19th December 2020), there are 76,186,444 confirmed COVID-19 cases worldwide, with 1,684,864 deaths, according to WHO. It shows that the global mortality rate is 6.9%. The USA’s confirmed COVID-19 cases are 17,899,267 and 321,025 deaths, which is top of the list in the world. Confirmed COVID-19 cases in India are 10,013,478, Brazil are 7,163,912, Russia are 2,819,429, France are 2,442,990, Turkey are 1,982,090, and UK are 1,977,167, respectively. In these countries, the number of deaths are 145,298, 185,687, 50,347, 60,229, 17,610, and 66,541, respectively to date (19th December 2020). And these cases are increasing day by day. Italy is another highly affecting country by this virus, and the positive reported cases are 1,921,778, and total deaths are 67,894. The Asian countries such as India is on the top, which is highly affected by this virus. In Pakistan, this rate is much slower as compared to other Asian Countries.
COVID-19 poses a major healthcare problem worldwide [
The rest of the manuscript is organized as follows: The existing relevant studies are discussed in Section 2 (related work). The proposed methodology is described in Section 3, which includes dataset preparation, features fusion, and selection. Section 4 represents the experimental results, and finally, analysis and conclusion are presented in Section 5.
Recently, COVID-19 patients have been diagnosed by reference to their X-ray and CT images using computer vision (CV)-based machine-learning algorithms, principally, supervised learning and deep learning (DL) techniques. Apostolopoulos et al. [
We develop automated prediction of positive COVID-19 pneumonia cases using CT scans. The positive cases are labeled via RT-PCR testing. The scheme features four steps. First, a database was prepared by collecting COVID-19 pneumonia-positive chest CT scans and normal scans from the Radiopaedia COVID-19 website (
All images (58 patients) were collected from the Radiopaedia COVID-19 website. We downloaded data on 30 patients with COVID-19 pneumonia confirmed via RT-PCR. We gathered 3,000 COVID pneumonia-positive images and 2,500 control images (
In terms of pattern recognition and machine-learning, features play pivotal roles in object representation. Many feature selection techniques for medical images are available [
Dynamic Average LBP (DALBP) features are modifications of the original LBP features; averages are used rather than central values. As in the original LBP [
where
As the image dimensions are
where
Consider an image of dimensions
where
where
where the length of each final feature vector [
SFTA features, also termed textural features, are also used to extract discriminatory information. In the medical context, SFTA features are often used to describe organs. The principal textural descriptor is the GLCM, but the use thereof is very time-consuming. We used accurate SFTA features that can be rapidly extracted [
where
where
Discrete Wavelet Transform (DWT) is a well-known type of feature extraction that analyzes images at various scales and resolutions [
where
where
The Renyi entropy feature vector is computed in row coefficients; the output vector is of dimensions
All extracted features were fused employing canonical correlation analysis (CCA) [
where
Next, Lagrange multipliers are used to solve the maximization problem between
Finally, the transformed features are combined as follows:
where
Feature selection involves selection of the best subset of input feature vectors based on a defined criterion. Use of a best subset improves learning and predictive accuracy, and reduces the computational time. We implemented a GA and an ELM fitness function. Most researchers use the Fine KNN and SVM for fitness calculations; however, we believe that the ELM is more efficient. The GA is Algorithm 1. The initial population size is 100, the number of iterations 1,000, the crossover rate (
Given a robust feature vector after GA and appropriate labeling
where
where
Based on this error,
We used publicly available chest CT images (Section 3.1). We extracted DALBP, SFTA, and DWT-plus-Entropy features, fused them using the CCA approach, employed a GA to select robust features, and delivered these to the ELM classifier for final predictions. We compared ELM performance to those of Logistic Regression, Q-SVM, Fine Gaussian, Fine Tree, and Cubic KNN in terms of accuracy, Precision, Recall, Specificity, and the F1 score. We used an Intel Core i7 8th generation CPU equipped with 16 GB of RAM and 8 GB of GPU running MATLAB 2019b software. We took a 60–40 approach with 10-fold cross-validation during both training and testing.
The predictions are presented in numerical form and as bar graphs. We explored the accuracies afforded by DALBP, SFTA, and DWT-plus-Entropy features; CCA fusion accuracy; and GA accuracy. For the DALBP features (
Classifier | Features bases accuracy | ||
---|---|---|---|
DALBP | SFTA | DWT-Entropy | |
ELM | 82.30 | 81.62 | |
Linear SVM | 82.42 | 75.27 | 80.09 |
Naïve Bayes | 81.49 | 73.78 | |
Quadratic SVM | 71.62 | 77.51 | |
Cosine KNN | 80.20 | 71.96 | 79.43 |
Medium KNN | 81.21 | 77.52 | 78.63 |
ESDA | 80.14 | 75.06 | 79.61 |
MG SVM | 81.63 | 72.94 | 80.33 |
Logistic regression | 79.38 | 77.52 | 78.41 |
Fine tree | 81.52 | 74.22 | 79.08 |
Fine Gaussian SVM | 78.12 | 74.36 | 77.92 |
Cubic KNN | 79.68 | 79.42 | 77.51 |
Cubic SVM | 80.72 | 73.21 | 75.04 |
EBT | 76.71 | 76.13 | 75.91 |
The CC-based fusion approach was then employed for prediction. The ELM classifier was best: 92.8% accuracy, 93.81% precision, 94% specificity, and an F1 score of 0.93. The worst classifier was the EBT (86.1% accuracy). All results are shown in
Classifier | Accuracy (%) | Precision (%) | Recall (%) | Specificity (%) | AUC | F1 score (%) |
---|---|---|---|---|---|---|
ELM | 92.5 | 1.00 | ||||
Linear SVM | 92.7 | 90.57 | 93.0 | 90.0 | 0.98 | 91.77 |
Naïve Bayes | 92.6 | 90.48 | 92.5 | 90.0 | 0.96 | 91.48 |
Quadratic SVM | 92.2 | 92.00 | 92.0 | 92.0 | 0.99 | 92.00 |
Cosine KNN | 92.2 | 90.48 | 92.5 | 90.0 | 0.96 | 91.48 |
Medium KNN | 91.5 | 90.29 | 91.5 | 90.0 | 1.00 | 90.89 |
ESDA | 90.8 | 89.42 | 91.0 | 89.0 | 0.99 | 90.20 |
MG SVM | 90.6 | 89.32 | 90.3 | 90.5 | 0.99 | 91.60 |
Logistic regression | 90.3 | 91.75 | 90.5 | 92.0 | 1.00 | 91.12 |
Fine tree | 90.2 | 90.82 | 90.0 | 91.0 | 1.00 | 90.41 |
Fine Gaussian SVM | 90.2 | 90.00 | 90.0 | 90.0 | 0.95 | 90.00 |
Cubic KNN | 89.8 | 91.67 | 90.0 | 92.0 | 0.95 | 90.83 |
Cubic SVM | 89.7 | 89.90 | 89.5 | 90.0 | 0.99 | 89.70 |
EBT | 86.1 | 91.95 | 86.5 | 93.0 | 0.92 | 89.14 |
We used an improved GA to select the best features for final prediction (Section 3.4 and Algorithm 1). The predictive performances of several classifiers are shown in
Classifier | Accuracy (%) | Precision (%) | Recall (%) | Specificity (%) | AUC | F1 score (%) |
---|---|---|---|---|---|---|
ELM | 1.00 | |||||
Linear SVM | 93.4 | 92.03 | 93.6 | 92.0 | 1.00 | 92.81 |
Naïve Bayes | 93.7 | 93.05 | 93.9 | 94.0 | 0.98 | 93.47 |
Quadratic SVM | 93.6 | 93.07 | 93.5 | 93.0 | 0.99 | 93.28 |
Cosine KNN | 93.6 | 93.07 | 93.5 | 93.0 | 0.99 | 93.28 |
Medium KNN | 93.5 | 93.07 | 93.5 | 93.0 | 1.00 | 93.28 |
ESDA | 93.4 | 92.16 | 93.0 | 92.0 | 0.99 | 92.58 |
MGSVM | 93.4 | 92.23 | 93.5 | 92.0 | 0.99 | 92.86 |
Logistic regression | 93.4 | 93.12 | 93.5 | 94.0 | 1.00 | 93.31 |
Fine tree | 93.2 | 89.19 | 93.5 | 88.0 | 0.98 | 91.29 |
Fine Gaussian SVM | 93.0 | 92.16 | 93.0 | 92.0 | 0.98 | 92.58 |
Cubic KNN | 92.9 | 92.16 | 93.0 | 92.0 | 1.00 | 92.58 |
Cubic SVM | 92.8 | 92.16 | 93.0 | 92.0 | 0.98 | 92.58 |
EBT | 89.5 | 90.72 | 89.5 | 91.0 | 0.94 | 90.11 |
Classifier | Min accuracy (%) | Average accuracy (%) | Max accuracy (%) | SEM (%) |
---|---|---|---|---|
ELM | ||||
LSVM | 92.24 | 92.82 | 93.40 | 0.4101 |
Naïve Bayes | 92.49 | 93.09 | 93.70 | 0.4277 |
Quadratic SVM | 92.38 | 92.99 | 93.60 | 0.4313 |
It is clear that fusion of multi-type features is valuable. This increases the number of predictors and enhances predictive accuracy. However, a few irrelevant features were added; if these are removed, accuracy is not compromised. Removal was effected via feature selection. We used an improved GA and an ELM to select the best features and improve predictive accuracy. In future work, we will seek a more efficient feature selection algorithm to improve accuracy further. Moreover, we will seek to build a larger image dataset that we will use to train a CNN.