Multithreshold Segmentation and Machine Learning Based Approach to Differentiate COVID-19 from Viral Pneumonia

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Coronavirus disease (COVID-19) has caused global devastation, necessitating efficient diagnostic tools.
• Chest X-ray (CXR) imaging offers valuable data for identifying localized infection regions.
• Automated screening and diagnosis using CXR can significantly aid healthcare systems.

Purpose of the Study:

• To develop a simple, threshold-based segmentation approach for identifying infection in CXR images.
• To investigate various texture features (intensity-based, wavelet transform, Laws) for differentiating COVID-19 from viral pneumonia (VP).
• To implement machine learning classifiers for automated COVID-19 diagnosis.

Main Methods:

• A threshold-based segmentation approach was used to detect potential infection areas in CXR images.
• Intensity-based, wavelet transform (WT)-based, and Laws-based texture features were extracted and analyzed.
• Feature selection was performed using Random Forest (RF), followed by classification using Support Vector Machine (SVM) and RF models.

Main Results:

• Intensity and WT-based features showed significant variations between COVID-19 and VP.
• Combined features, trained with SVM and RF classifiers, effectively differentiated between the two conditions.
• The RF model achieved a high classification accuracy of 0.9 and an Area Under the Curve (AUC) of 0.97.

Conclusions:

• The implemented methodology demonstrates utility in characterizing and differentiating COVID-19 from VP using CXR.
• Automated analysis of CXR features shows promise for rapid and accurate diagnosis.
• Machine learning models, particularly RF, can effectively support clinical decision-making in respiratory infections.