European Radiology: AI-based correction of brain CT motion artifacts

As a rapid imaging method, computed tomography (CT) of the brain can non-invasively obtain information about brain tissue and brain structure, and has been widely used in the clinical diagnosis
of various brain diseases.
In settings such as emergency departments, patients are often prone to uncontrolled head movements, leading to motion artifacts that affect the diagnostic accuracy
of CT images.
Therefore, this becomes an important quality limiting factor
for brain CT imaging.
How to minimize motion artifacts in brain CT to reduce repeated scans and additional radiation is the focus
of clinical attention at this stage.

Recently, artificial intelligence (AI) has shown excellent clinical value and promise in the field of medical imaging, helping with lesion segmentation, disease detection, image reconstruction and motion artifact correction
.
To reduce motion artifacts in cranial CT images without the need for rescans or additional motion tracking systems, Su et al.
recently reported an AI-based MC reconstruction algorithm that uses multi-scale convolutional neural networks (CNNs) to learn and reduce motion artifacts.

However, the study only focused on proof-of-concept and technical efficiency, while clinical manifestations remain unknown
.

Recently, a study published in the journal European Radiology verified the clinical application effect of the previously proposed artificial intelligence-based MC algorithm in clinical cranial CT applications, which provides support
for clinically effective inhibition of motion artifacts, improvement of overall image quality, and improvement of diagnostic confidence of cranial CT.

A total of 53 cases were included in the study, and each patient was rescanned
for motion artifacts found on the first scan.
The rescanned image is reconstructed using the mixed iterative reconstruction (IR) algorithm (reference group), while the image scanned for the first time is reconstructed
with the mixed IR (motion group) and MC algorithm (MC group).
Image quality
is compared in terms of standard deviation (SD), signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), mean squared error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM) and mutual information (MI), and subjective scores.
The diagnostic performance of each case was accordingly assessed
by lesion detectability or early CT score (ASPECTS) assessment of the Alberta Stroke Program.

Compared to the motion artifacts group, SNR and CNR were significantly increased
in the MC group.
Compared to the reference group, MSE, PSNR, SSIM and MI improved by 44.
1%, 15.
8%, 7.
4% and 18.
3%, respectively (all P<0.
001).

The subjective image quality index scored higher in the MC group than in the sports group (P < 0.
05).

In the ASPECTS assessment, lesions were more detectable in the MC group than in the exercise group and the AUC was higher (0.
817 vs 0.
614).

Figure Comparison
of reference images, moving images, and corrected images.
SSIM, structural similarity index; PSNR, peak signal-to-noise ratio; MSE, mean squared error; MI, mutual information

In summary, this study has been clinically verified
by an AI-based cranial CT reconstruction algorithm.
The algorithm demonstrates excellent ability to reduce motion artifacts and improve image quality, helping to improve the success rate of cranial CT examinations, avoid unnecessary rescans, increase diagnostic confidence for radiologists, and ultimately accelerate the workflow
of cranial CT scans in emergency departments.

Original source:

Leilei Zhou,Hao Liu,Yi-Xuan Zou,et al.
Clinical validation of an AI-based motion correction reconstruction algorithm in cerebral CT.
DOI:10.
1007/s00330-022-08883-4