AI algorithm optimizes CCTA image quality

Deep learning can improve the quality of coronary CT angiography (CCTA) images, according to a new study published in Academic Radiology.

“CCTA is a reliable and accurate modality to detect and exclude coronary artery disease (CAD) due to its high sensitivity and negative predictive value,” wrote lead author Peijun Liu, MD, Chinese Academy of Medical Sciences and Peking Union Medical College in Beijing, and colleagues. “However, the potential risks of radiation-induced carcinogenesis and contrast-induced nephropathy have attracted attention.”

Numerous steps have been taken to decrease the radiation dose required with CCTA exams and reduce those potential risks—but this has resulted in diminished image quality (IQ). The study’s authors aimed to address and evaluate a deep learning-based algorithm’s ability to reduce the image noise associated with CCTA, improving the IQ of each image.

Liu’s team studied data from 70 consecutive patients with suspected or known CAD who were treated from May to July 2018. Participants were separated into two groups of 35 patients, with patients from group A undergoing CCTA with 80 kVp and patients from group B undergoing CCTA with 100 kVp. A third group of patients, group C, was created when the team’s deep learning algorithm was applied to the data from group A.

Researchers then compared the IQ of these three groups, focusing on image noise, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) for an objective analysis and bringing in two experienced radiologists for a more subjective analysis.

Overall, the optimized images found in group C had decreased image noise, increased SNR and increased CNR compared to the unoptimized images found in group A. In addition, the subjective analysis found that group C had “significantly better” IQ than group A. The IQ of groups B and C were “not significantly different.”

“This study demonstrates that the application of DL-based optimization algorithm could effectively reduce the image noise of CCTA to improve IQ with 80 kVp while achieving the goal of low radiation dose and contrast volume,” the authors wrote. “After application of the DL-based optimization algorithm, the image quality of prospectively ECG-triggered CCTA at 80 kVp is comparable to that at 100 kVp with iterative reconstruction.”

Liu et al. did note that their study had certain limitations. The sample size was just 70 patients, for example, and “diagnostic accuracy was not evaluated by invasive coronary angiography, which is considered the gold standard.”