A ensemble methodology for automatic classification of chest X-rays using deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Chest X-rays are vital for diagnosing numerous medical conditions.
• Accurate interpretation of chest X-rays is crucial for patient care.
• Current diagnostic methods can be time-consuming and require expert interpretation.

Purpose of the Study:

• To develop and evaluate a Convolutional Neural Networks (CNNs) ensemble model for automated chest X-ray classification.
• To improve the efficiency and accuracy of chest X-ray screening.
• To categorize chest X-rays into High Confidence Normal (HCn), Borderline classification (BC), or High Confidence Abnormal (HCa).

Main Methods:

• Utilized a private dataset of frontal and lateral projection chest X-ray images.
• Employed an ensemble of VGG-16, ResNet50, and DenseNet121 CNN architectures.
• Implemented a Confidence Threshold (CTR) for classifying predictions.

Main Results:

• Achieved 54.63% of exams classified with high confidence.
• Classified 32% of normal exams as High Confidence Normal (HCn) with a 1.68% False Discovery Rate (FDR).
• Classified 23% of abnormal exams as High Confidence Abnormal (HCa) with a 4.91% False Omission Rate (FOR).

Conclusions:

• The CNN ensemble methodology shows significant promise for aiding chest X-ray diagnosis.
• High confidence classification rates suggest potential for efficient screening.
• The model's performance indicates a valuable tool for radiologists, reducing workload and improving diagnostic accuracy.