CXray-EffDet: Chest Disease Detection and Classification from X-ray Images Using the EfficientDet Model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiography

Background:

• Machine learning shows promise in clinical tasks, especially medical imaging analysis.
• Chest radiography is crucial for diagnosing anomalies, but complex image structures and artifacts pose challenges.
• Automated detection of chest abnormalities from X-rays is an ongoing area of research.

Purpose of the Study:

• To address the complexities in chest X-ray analysis for reliable disease detection and classification.
• To propose a novel deep learning (DL) approach for identifying chest abnormalities using X-ray images.

Main Methods:

• Developed the EfficientDet (CXray-EffDet) model, utilizing EfficientNet-B0-based EfficientDet-D0.
• Employed the model for feature computation, detection, and classification of eight categories of chest abnormalities.
• Validated the model on a large dataset from the National Institutes of Health (NIH).

Main Results:

• The CXray-EffDet model demonstrated high recall and computational robustness.
• Achieved an Area Under the Curve (AUC) score of 0.9080.
• Obtained an Intersection over Union (IOU) of 0.834 for localization and categorization performance.

Conclusions:

• The proposed CXray-EffDet model effectively enhances chest abnormality recognition.
• The model's performance indicates its competency in localizing and categorizing chest diseases from X-ray images.
• This deep learning approach offers a lightweight and computationally robust solution for medical imaging analysis.