MS-ANet: deep learning for automated multi-label thoracic disease detection and classification.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Chest X-rays are crucial for diagnosing chest diseases.
• Automatic classification of radiological images aids clinical decisions.
• Challenges include disease-specific receptive fields and data imbalance.

Purpose of the Study:

• To propose a multi-label chest disease image classification scheme.
• To address challenges in classifying chest diseases from X-rays.
• To enhance diagnostic accuracy and radiologist efficiency using AI.

Main Methods:

• Developed a multi-scale attention network for iterative information fusion.
• Focused on disease-probability regions to extract meaningful data.
• Designed a novel loss function for improved visual perception and classification consistency.
• Conducted experiments on Chest X-Ray14 and CheXpert datasets.

Main Results:

• Achieved state-of-the-art Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.850 on Chest X-Ray14.
• Attained an AUROC of 0.815 on the CheXpert dataset.
• Demonstrated the effectiveness of the proposed method in chest X-ray image classification.

Conclusions:

• The multi-scale attention network effectively classifies chest diseases from X-rays.
• The AI algorithm shows significant potential in assisting radiologists.
• This approach can improve diagnostic accuracy and work efficiency in clinical settings.