Advancing chest X-ray diagnostics: A novel CycleGAN-based preprocessing approach for enhanced lung disease...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Chest radiography is crucial for diagnosing thoracic diseases.
• Image artifacts like wires and electrodes introduce noise, hindering accurate diagnosis.
• Artifacts obscure anatomical structures in X-ray images.

Purpose of the Study:

• To develop a preprocessing approach for artifact detection and noise reduction in chest X-rays.
• To improve the accuracy of lung disease classification models.
• To enhance the distinction of relevant anatomical structures in medical images.

Main Methods:

• A novel artifact detection and noise reduction preprocessing approach using CycleGAN for image sharpening.
• Implementation of DenseNet-121 with channel and spatial attention mechanisms for classification.
• Integration of clinical characteristics alongside image data for enhanced model performance.

Main Results:

• The preprocessing approach significantly improved classification performance.
• Area Under the Curve (AUC) increased by 5.91% for pneumonia and 6.44% for consolidation.
• The method outperformed previous studies on the ChestX-Ray14 dataset for 14 diseases.

Conclusions:

• Artifact removal is critical for accurate lung disease diagnosis from radiographic images.
• The proposed method allows AI models to focus on relevant diagnostic features, boosting performance.
• This artifact preprocessing technique shows promise for wider applications in medical image analysis.