A Pneumonia Diagnosis Scheme Based on Hybrid Features Extracted from Chest Radiographs Using an Ensemble Learning...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Pneumonia is a leading cause of child mortality globally, particularly in developing nations lacking timely diagnostic tools.
• Accurate and rapid diagnosis of pneumonia is crucial for effective treatment and reducing mortality rates.
• Machine learning (ML) offers potential for cost-effective, early, and automated pneumonia detection.

Purpose of the Study:

• To develop and evaluate a novel ML-based method for detecting pneumonia and differentiating between bacterial and viral types using chest radiographs.
• To enhance diagnostic capabilities in resource-limited settings through automated analysis of medical images.

Main Methods:

• A three-class classification approach was employed using chest X-ray images.
• Data augmentation was used to balance dataset sample sizes.
• Statistical and deep learning-derived global features were extracted, combined, and fed into a RandomForest classifier.
• Feature selection was applied to identify the most relevant features for classification.

Main Results:

• The proposed ML model achieved 86.30% classification accuracy and an 86.03% F-score on a chest radiograph dataset.
• The model demonstrated efficacy and reliability in classifying pneumonia cases.
• A notable challenge remains in accurately distinguishing between viral and bacterial pneumonia subtypes.

Conclusions:

• The developed ML method provides a promising, fast, and automated approach for pneumonia detection from chest X-rays.
• While effective for general pneumonia detection, further refinement is needed to improve the differentiation between bacterial and viral pneumonia.
• This technology has the potential to significantly improve diagnostic accessibility and speed in clinical settings.