Denoising Multiphase Functional Cardiac CT Angiography Using Deep Learning and Synthetic Data.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary CT angiography (CCTA) is vital for cardiac diagnosis.
• Dose modulation reduces radiation exposure but increases image noise, hindering functional analysis.
• Existing noise reduction methods may not adequately preserve functional image quality.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) method for denoising functional cardiac imaging derived from CCTA.
• To assess the performance of DL-based denoising against standard algorithms and unprocessed images.
• To validate the utility of DL-denoised images for accurate cardiac function quantification.

Main Methods:

• A retrospective study utilizing 566 CCTA datasets.
• Development of a 3D convolutional neural network for multiphase image denoising.
• Comparison of DL denoising with block-matching and 3D filtering (BM3D) and unprocessed images.
• Quantitative analysis of noise, signal-to-noise ratio, and expert image quality assessment.
• Validation using threshold-based segmentation and comparison with manual measurements.

Main Results:

• DL-based denoising significantly reduced image noise compared to BM3D (SD of left ventricular blood pool: 20.3 HU vs 33.4 HU; P < .0001).
• Expert evaluations showed superior image quality for DL-denoised images over standard denoising.
• Semiautomatic measurements on DL-denoised images highly correlated with manual measurements (ICC, 0.97).

Conclusions:

• 3D deep learning effectively denoises functional cardiac imaging from CCTA.
• DL-based denoising enhances image quality and facilitates accurate cardiac function analysis.
• This method supports valid automatic processing of cardiac functional imaging with reduced radiation dose.