Anatomy-aware disease severity detection in chest X-ray images

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology AI Research

Background:

• Automatic disease detection in chest radiographs is advancing.
• Current AI efforts often overlook disease severity assessment.
• Severity is a key component of radiological findings.

Purpose of the Study:

• To develop an AI model for accurate disease severity classification in chest radiographs.
• To utilize anatomy-aware severity labels from the Chest ImaGenome dataset.
• To address the gap in AI-driven severity determination.

Main Methods:

• Proposed a multitask architecture for representation learning.
• Employed a detection network for region-specific feature extraction.
• Integrated a self-attention module for feature refinement.
• Enabled simultaneous disease and severity classification.

Main Results:

• Achieved an area under the ROC curve of 0.79 for severity detection.
• Demonstrated the effectiveness of anatomy-aware labels.
• The model provides anatomy-aware disease and severity labels.

Conclusions:

• The proposed multitask architecture effectively classifies disease severity in chest radiographs.
• Anatomy-aware severity classification is feasible and crucial.
• This work advances AI applications in diagnostic radiology.