Deep learning automates head CT reformatting, bolstering workflow efficiencies
Deep learning can help automate the arduous task of reformatting CT exams, bolstering workflow efficiencies in the process, according to new research.
Head computed tomography is a fundamental diagnostic modality for many neurological conditions. However, conventional manual processes for generating reformats are susceptible to variability, due to factors such as patient positioning, comfort and individual technologist preferences, experts write in JACR.
This process, which involves creating new images from the initial 2D slices, also can introduce diagnostic errors, along with delaying turnaround times and taking up valuable resources. Researchers with the University of California, Irvine, recently experimented with using artificial intelligence to automate the reformatting process.
They’ve demonstrated early success, generating expert-level CT reformats with high accuracy and consistency.
“The implementation of such an automated method offers the potential to improve standardization, increase workflow efficiency and reduce operational costs in clinical practice,” Peter D. Chang, MD, with UC Irvine’s Department of Radiological Sciences, and co-authors concluded.
For the study, researchers utilized a single-shot foundation model previously trained to localize any anatomical structure in the body on CT without requiring task-specific tuning. The procedure for reformat generation starts with the user defining a field of view with custom rotation, centering and zoom parameters, the authors noted. Next, the operator then identifies a set of landmarks that define the reformat plane and field of view. With any given exam, the foundation model helps identify the relevant landmarks, generating the corresponding reformat through landmark-based image alignment.
The AI model was used to create automated reformats for nearly 1,800 consecutive, noncontrast head CT exams. Results from AI were used as a reference standard to evaluate quality and speed against original, tech-generated reformats from the dataset. Deep learning was able to generate expert-level CT formats with high accuracy and consistency, achieving less than 1 degree of rotational error, along with less than 1% centering and zoom error. Chang and colleagues noted significant variability in manual reformatting quality, observed across different factors including patient age, scanner location, report findings and individual techs.
“The persistence of such wide variation even among high-volume operators (only those with over 100 scans in the study period) suggests that experience alone may not be sufficient to ensure consistent reformat quality and efficiency,” the authors reported. “This data further supports the critical importance and value of an AI-automated solution to mitigate human variability and ensure standardized, high-quality head CT reformats across all examinations.”
Read more, including potential study limitations, in the Journal of the American College of Radiology.
Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic.