Equilibrium-based COVID-19 diagnosis from routine blood tests: A sparse deep convolutional model

Abstract

SARS-CoV2 (COVID-19) is the virus that causes the pandemic that has severely impacted human society with a massive death toll worldwide. Hence, there is a persistent need for fast and reliable automatic tools to help health teams in making clinical decisions. Predictive models could potentially ease the strain on healthcare systems by early and reliable screening of COVID-19 patients which helps to combat the spread of the disease. Recent studies have reported some key advantages of employing routine blood tests for initial screening of COVID-19 patients. Thus, in this paper, we propose a novel COVID-19 prediction model based on routine blood tests. In this model, we depend on exploiting the real dependency among the employed feature pool by a sparsification procedure. In this sparse domain, a hybrid feature selection mechanism is proposed. This mechanism fuses the selected features from two perspectives, the first is Pearson correlation and the second is a new Minkowski-based equilibrium optimizer (MEO). Then, the selected features are fed into a new 1D Convolutional Neural Network (1DCNN) for a final diagnosis decision. The proposed prediction model is tested with a new public dataset from San Raphael Hospital, Milan, Italy, i.e., OSR dataset which has two sub-datasets. According to the experimental results, the proposed model outperforms the state-of-the-art techniques with an average testing accuracy of 98.5% while we employ only less than half the size of the feature pool, i.e., we need only less than half the given blood tests in the employed dataset to get a final diagnosis decision.

Keywords: 1DCNN; Blood tests; COVID-19; Equilibrium optimization; Feature pool sparsification; Feature selection; Pearson correlation.

Fig. 1

A logarithmic scale for COVID-19 Monthly total deaths (Worldometer, 2020).

Fig. 2

An illustration of the proposed COVID-19 prediction model.

Fig. 3

COVID-19 examination results for COVID-specific dataset in (a) and for CBC dataset in (b).

Fig. 4

COVID-19 swab result distribution according to age and gender for COVID-specific dataset in (a) and for CBC dataset in (b).

Fig. 5

3D Visualization of the predicted outliers/inliers in COVID-specific dataset via three PCA components.

Fig. 6

An illustration of the proposed feature selection technique that is based on a fusion process between Pearson dropping (PCC) and the introduced Minkowski-based equilibrium optimizer (MEO) in a serial and parallel manner in the original features domain $\mathcal{F}$ once and in the proposed sparse domain ${\mathcal{F}}_S$ another. $"+"$ represents combining decisions by OR operations while $"\times "$ represents seeking the intersections of decisions by AND operations.

Fig. 7

Pairwise Pearson correlation of features: (a), (c) for the original feature pool $\mathcal{F}$ while (b), (d) for the sparsified feature pool ${\mathcal{F}}_S$. The first row for COVID-specific dataset and the second one for CBC dataset.

Fig. 8

2D illustration of Equilibrium candidates’ collaboration in updating particles’ concentration.

Fig. 9

Flow chart of the proposed MEO algorithm.

Fig. 10

Comparison of the results of average fitness over iterations for the traditional EO, in the first row, and the proposed MEO, in the second one, for COVID-specific dataset. The first column is the results of the original feature pool $\mathcal{F}$ while the second one for the sparsified feature pool ${\mathcal{F}}_S$.

Fig. 11

An example of 1DCNN model for a binary classification problem. In this example, the network consists of two convolutional layers (Conv_1with 32 filters and Conv_2 with 64 filters), Max pooling layer, flattening layer and finally some fully connected layers with soft-max layer.

Fig. 12

Summary of the proposed 1DCNN for COVID-19 prediction considering 9 selected features.

Fig. 13

The employed evaluation metrics.

Fig. 14

Confusion matrices of testing the proposed COVID-19 prediction algorithm adopting the four cases indicated in Table 4 showing the effect of SMOTE and iForest on the performance.

Fig. 15

AdaBoost feature importance employing all features for COVID-specific dataset in the first row and CBC dataset in the second row. (a) and (c) in the features original domain while (b) and (d) in the sparse domain.

Fig. 16

AdaBoost feature importance, for COVID-specific dataset, adopting the followings: 1. PCC-based feature selection in the features original domain (22 feature selected) while applying training and testing once for the original samples in $\mathcal{F}$ (a), and another for the sparse samples in ${\mathcal{F}}_s$ (b). 2. PCC-based feature selection in sparse domain (14 feature selected) while applying training and testing once for the original samples in $\mathcal{F}$ (c), and another for the sparse samples in ${\mathcal{F}}_s$ (d). 3. MEO-based feature selection in features original domain (12 feature selected) while applying training and testing once for the original samples in $\mathcal{F}$ (e), and another for the sparse samples in ${\mathcal{F}}_s$ (f). 4. PCC-based feature selection in sparse domain (12 feature selected) while applying training and testing once for the original samples in $\mathcal{F}$ (g), and another for the sparse samples in ${\mathcal{F}}_s$ (h).

Fig. 17

Classification reports of testing the proposed COVID diagnosis model based on the proposed fused selection method and 1DCNN in both original domain (a), (c) and sparse domain (b), (d). The first two rows belong to COVID-specific dataset while the other rows belong to CBC dataset.

Fig. 18

Training-validation performance in terms of accuracy for the proposed COVID prediction algorithm. The first row for COVID-specific dataset and the other one for CBC dataset. The training in (a), (c) is performed in features original domain and the others (b), and (d) in sparse domain. The training is performed over the selected features by the proposed fused-based feature selection mechanism which results 13 features for COVID-specific dataset and 6 features for CBC-dataset.

Fig. 19

Classification reports of testing the following studies: (Alakus & Turkoglu, 2020) {18/33–13/13} as (a), (AlJame et al., 2020) {18/33–13/13} as (b), (Cabitza et al., 2021) {33/33–13/13} as (c), (Brinati et al., 2020) {33/33–13/13} as (d), (Shaban et al., 2021) {33/33–13/13} as (e), and Ours {13/33–6/13} as (f). {} denotes {selected features/total no. of features for COVID-specific dataset – selected features/total no. of features for CBC dataset. (?.1) for COVID-specific dataset and (?.2) for CBC dataset.