AI shows high accuracy for 3D upper airway segmentation in CBCT/CT scans versus manual methods
This systematic review and meta-analysis evaluated the efficacy of artificial intelligence (AI) tools, including deep learning and machine learning, for 3D segmentation of the upper airway in cone-beam computed tomography (CBCT) or CT scans. The analysis included 11 studies, with 6 contributing to meta-analyses, comparing AI-based segmentation against manual analysis performed by human experts. The primary outcome was the analysis of upper airway images, with secondary metrics including accuracy, precision, dice similarity scores, and total volume differences.
Results from the meta-analyses indicated that AI tools achieved performance metrics above 90% for precision, dice similarity score, intersection over union, and recall. The total volume difference between AI and manual segmentation was reported as small but significantly above zero. The review's qualitative assessment also described promising results for AI segmentation efficacy. Most included studies were assessed as having a low risk of bias, and sensitivity analyses supported the robustness of the meta-analysis findings, with no significant publication bias detected.
Key limitations noted by the authors include the need for many more studies before decisive conclusions can be drawn. Safety and tolerability data for the AI tools were not reported. The evidence demonstrates an association between AI use and high segmentation accuracy in this technical, image-analysis context. For clinical practice, these findings highlight the potential for AI to serve as an assistive tool in radiological and dental imaging workflows, but they do not yet support replacing expert manual segmentation.
View Original Abstract ↓
Muscle ultrasonography shows high diagnostic accuracy for fasciculation detection in ALS
A Simple Scan Could Speed Up an ALS Diagnosis
Related in Radiology & Imaging
From Other Specialties
