Noncontrast head CT with AI helps to localize hemorrhage: Study
USA: The use of noncontrast head CT (NCCT) scans with an artificial intelligence (AI) algorithm may help clinicians to better identify, localize, and characterize intracranial hemorrhages (ICHs), says a recent study. The study was published in the the journal Radiology: Artificial Intelligence on April 20, 2022.
The retrospective study was conducted by Eli Gibson and colleagues to present a method that automatically detects, subtypes and locates acute/subacute ICH on NCCT and generates detection confidence scores to detect high-confidence data subsets with higher accuracy and improve radiologic worklist prioritization. Such score warrant clinicians for better use of AI tools.
The study included 46,057 studies from seven 'internal' centers for development (training/architecture selection/hyperparameter tuning/operating point calibration) (n = 25,946) and evaluation (n = 2,947) and three 'external' centers for calibration (n = 400) and evaluation (n = 16,764).
Development data was contributed by 'Internal' centers while 'external' centers did not. ICH and subtype presence (intraparenchymal/intraventricular/subarachnoid/subdural/epidural) and segmentations per case was predicted by deep neural networks. Two ICH confidence scores are discussed: a calibrated classifier (CC) entropy, and a Dempster-Shafer score (DS).
The study led to the following findings:
· The area-under-the-curve for ICH was 0.97 [0.97, 0.98] and 0.95 [0.94, 0.95] on internal and external center data, respectively.
· On 80% of the data stratified by CC & DS scores, the system improved the Youden's index for internal centers from 0.84 to 0.93 (CC) and 0.92 (DS), respectively, and for external centers from 0.78 to 0.88 (CC) and 0.89 (DS).
· Models estimated 27% (CC) and 27% (DS) and 25% (CC) and 27% (DS) decreases, respectively, in RTAT for AI-prioritized worklists with versus without confidence measures.
The authors concluded, "NCCT ICH detection with statistical confidence reliably detected and subtyped hemorrhage, identified high-confidence predictions, and improved worklist prioritization in simulation."
Reference:
The study titled, "AI with Statistical Confidence Scores for Detection of Acute/Subacute Hemorrhage in Noncontrast Head CT Scans," was published in the journal Radiology: Artificial Intelligence.
