Harvard radiology researchers have demonstrated the use of AI to instantly recognize motion artifacts likely to be problematic for readers of certain chest CT scans.
The capability could allow immediate alerting of CT technologists, who would adjust scan protocols or re-scan patients to optimize image quality prior to physician interpretation.
Importantly, the investigators note, the model’s primary front-end developer was a radiologist with subspecialist training in thoracic imaging.
Further, while the radiologists tested the AI for use in CT pulmonary angiography (CTPA), they suggest, albeit with caution, that it may have some potential for adaptation to other chest CT diagnostics.
The preprint server medRxiv has posted the study ahead of peer review [1].
Motion Artifacts, Respiratory Motion, Technically Inadequate—And So On
Lead author Giridhar Dasegowda, MBBS, senior author Keith Dreyer, DO, PhD, and colleagues at Mass General Brigham drew 793 CTPA reports from three care sites (two academic, one community). The queries yielding the results contained terms like motion artifacts, respiratory motion and technically inadequate.
To train the model, the team exported 554 of the 793 images (70%) to a commercial AI modeling prototype. They held out the remaining 239 images (30%) for validation.
The researchers established ground truth with input from a radiologist with 16 years of subspecialty experience in thoracic reads.
Challenging the experimental algorithm to classify each image as adequate or inadequate for confident diagnostic interpretation—i.e., “motion” or “no motion”—they found the model returned correct classifications with 94% sensitivity, 91% specificity, 93% accuracy and 0.93 area under the ROC curve.
In their discussion, the authors note that prior studies have investigated AI’s ability to identify anatomic regions with motion artifacts.
Their study, however, “uses a single coronal multiplanar reformatted image per CT exam and therefore could be more time-efficient, while ignoring region-specific, sparse motion artifacts which do not require repeat acquisition,” they write. “To our best knowledge, there are no peer-reviewed reports on the use of AI models for identifying motion artifacts in CTPA or chest CT examinations.”
Toward Mitigating ‘Substantial Motion Impairment on Diagnostic Evaluability’
Emphasizing that their model’s development stuck to CTPA diagnostics and thus might not apply to other chest CT protocols or body regions, Dasegowda et al. comment:
[Our] physician-trained and tested AI model can help identify substantial motion artifacts on CT pulmonary angiography. Automatic recognition of such artifacts can help CT technologists apply faster scan protocols and reacquire images to mitigate the impact of substantial motion impairment on diagnostic evaluability.”
In introducing their study report, the authors note that the chest represents one of the most frequently scanned body parts in CT but still “ranks high among the most challenging parts” to image with optimal diagnostic quality.
They also point out that motion artifacts are common with legacy scanners, critically ill patients and patients with shortness of breath and/or persistent cough.
The article preprint is available in full for free.
More Coverage of Imaging AI:
Work no less burdensome for 70% of 276 radiologists using AI
‘Vast and diverse’ repository of image data goes open-access for healthcare AI researchers
Seeds planted for a needs-based, radiology-specific AI curriculum
Ultrasound-embeddable AI sharp at diagnosing clogged carotid arteries
Image-based data commons enjoying fast growth thanks to RSNA, ACR, other contributing orgs
Radiologists best AI at diagnosing COVID on chest X-rays
Crowdsourced X-rays suitable for training AI in orthopedic injuries
4 increasingly common uses for imaging AI in the ED
Reference:
- Giridhar Dasegowda, Keith Dreyer, et al., “Auto-detection of motion artifacts on CT pulmonary angiograms with a physician-trained AI algorithm.” medRxiv, June 23, 2022. DOI: https://doi.org/10.1101/2022.06.23.22276818
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.