Nature-Inspired Algorithms-Based Optimal Features Selection Strategy for COVID-19 Detection Using Medical Images

In the case of communicable diseases, such as COVID-19, effective and quick testing techniques make it easier to identify a contaminated person, so that he or she can be easily isolated. To predict COVID-19-infected individuals through chest computed tomography scans, this study suggests an effective feature selection technique incorporated in clinical decision support system that may be used for testing. After pre-processing, we retrieved 213 features from the chest computed tomography images of a public data set with 2482 images. Then, in a two-step process, the most significant features for recognizing the difference between COVID patients and healthy individuals are selected. Initially, the Chi-square test selects 75% of the initial extracted features, which are then forwarded to three nature-inspired computing algorithms: the cuckoo search optimization algorithm, a teaching–learning-based optimization algorithm, and a hybrid of these two for further optimization. The finally selected reduced feature set and five machine learning classifiers are then employed to classify these computed tomography images. Twenty-four experiments using fivefold and tenfold cross-validation have been performed to find the best values for eight statistical efficiency evaluation metrics. Our suggested approach achieves a notable accuracy of 95.99%, the best mean intersection over union of 0.9655, and the highest area under curve of 0.9966. XGBoost delivers more effective, promising, and comparable results when compared to other ML classifiers. Our suggested testing approach will benefit frontline workers and the state by providing routine and cost-effective testing, and faster results.