X-ray2CTPA: leveraging diffusion models to enhance pulmonary embolism classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Chest X-rays (CXR) offer limited detail compared to CT scans.
• CT Pulmonary Angiography (CTPA) provides superior 3D data but is costly and involves higher radiation exposure.
• Bridging the gap between CXR and CTPA is crucial for accessible diagnostics.

Purpose of the Study:

• To develop a generative AI model for cross-modal translation from 2D CXR to 3D CTPA.
• To enhance diagnostic accuracy for conditions like Pulmonary Embolism (PE) using synthesized CTPA data.
• To explore the generalizability of the model for other medical imaging cross-modality translations.

Main Methods:

• A novel diffusion-based generative AI approach was employed for the X-ray to CTPA translation.
• Synthesized 3D CTPA images were integrated into a classification framework.
• Model performance was rigorously evaluated using quantitative metrics and a PE categorization task.

Main Results:

• The AI model successfully translated low-resolution 2D CXR to high-resolution 3D CTPA.
• Improved Area Under the Curve (AUC) was observed in the Pulmonary Embolism (PE) classification task.
• Generated images demonstrated diagnostic relevance, validated by quantitative assessments.

Conclusions:

• The proposed diffusion-based method enables high-fidelity cross-modal medical image translation.
• This approach offers a pathway towards more accessible and cost-effective advanced diagnostic imaging.
• The method shows promise for broader applications in medical imaging and diagnostics.