Multi-class classification of COVID-19 and other infections using machine learning model with wavelet and laws features

• Published in: AIP conference proceedings Vol. 2919; no. 1
• Main Authors: Basha, Shaik Mahaboob, Neto, Aloísio Vieira Lira, Elaziz, Mohamed Abd, Mohisin, Shaik Hashmitha, Albuquerque, Victor Hugo C. e
• Format: Journal Article Conference Proceeding
• Language: English
• Published: Melville American Institute of Physics 25.03.2024
• Subjects: Algorithms; Classifiers; Computed tomography; Coronaviruses; COVID-19; Diagnosis; Image classification; Image processing; Machine learning; Medical imaging; Physicians; Support systems; Texture; Viral diseases; Wavelet transforms
• ISSN: 0094-243X; 1935-0465; 1551-7616; 1551-7616
• DOI: 10.1063/5.0184741

Corona virus disease (COVID-19) was responsible for inevitable chaos in the human life due to its inherent nature of high contagious across the globe. This very reason also hindered the physicians and healthcare support systems to provide in-time medical aid to the suffering patients.The diagnosis of COVID-19 using clinical information is the prime source for detection and treatment. However, computerised tomography (CT) and Chest X-ray (CXR) images have found useful for screening purposes due to the successful usage of image based analysis by physicians which was aided by image processing based algorithms.Automated means of screening, diagnosis and prognostic studies using CT and CXR with the help of comprehensive image processing pipelines could invariably augment the efficiency in the respective tasks. The main objective of this study is to implement multi-class classification of COVID-19, viral-pneumonia (VP) and lung opacity (LO) using threshold-based segmented CXR images further transformed using wavelet transform (WT)and Laws based texture representation. Machine learning (ML) based implementation and comparison was carried using Random Forest (RF) and Logistic regression (LR) techniques with of the WT and Laws features individually as well as with combined feature vectors. The ML based implementation was aimed to classify the CXR image based on the individual or combined features into COVID-19, VP or LO pathology. The validation of the ML techniques was further carried using confusion matrix and other validation measures. The performance of the RF classifier was superior with WT features than Laws texture features on individual category of features than LR technique. Amarginal improvement was observed in the classifier performance when WT and Laws based features were combined.