Deep Learning Versus Iterative Reconstruction for CT Pulmonary Angiography in the Emergency Setting: Improved Image...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Computed tomography pulmonary angiography (CTPA) is crucial for diagnosing pulmonary embolism (PE).
• Traditional hybrid-iterative reconstruction (IR) techniques balance image quality and radiation dose.
• Advancements in deep learning-based image reconstruction (DLR) offer potential improvements.

Purpose of the Study:

• To compare the image quality and radiation dose of CTPA using a novel deep learning-based image reconstruction (DLR) algorithm versus the standard hybrid-iterative reconstruction (IR) technique.
• To evaluate quantitative and qualitative image quality metrics.
• To assess radiation dose parameters between the two reconstruction methods.

Main Methods:

• Retrospective review of 140 patients undergoing CTPA for suspected PE.
• Quantitative assessment included image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR).
• Qualitative assessment used a 5-point scale; radiation dose parameters (CTDIvol, DLP) were recorded.

Main Results:

• DLR images showed significantly lower noise and higher SNR and CNR compared to hybrid-IR (p < 0.01).
• DLR achieved significantly higher qualitative image quality scores for both soft and lung filters (p < 0.01).
• Radiation dose parameters (CTDIvol and DLP) were significantly lower with DLR reconstruction.

Conclusions:

• Deep learning-based image reconstruction (DLR) significantly improves CTPA image quality over hybrid-iterative reconstruction (IR).
• DLR enables a reduction in radiation dose while maintaining or enhancing image quality.
• DLR represents a promising advancement for CTPA examinations, offering better image quality and lower radiation exposure without compromising diagnostic confidence for PE.