Axial images at the L1 level, without (left) and with (right) segmentation overlay. Red indicates skeletal muscle, green indicates trabecular bone, yellow indicates visceral fat, and blue indicates subcutaneous fat. Segmented regions also include liver (beige) and spleen (orange), which were not assessed in the present analysis. (A) 78-year-old woman who underwent abdominal and pelvic CT at an external facility. The bone tool returned an L1 vertebral bone attenuation of -146 HU, outside the reference range. Therefore, the tool was considered a technical failure for the bone tool. The failure was attributed to the volume averaging of the vacuum phenomenon within the disk. (B) 64-year-old woman who underwent abdominal and pelvic CT at an external facility. The bone tool returned a vertebral body bone loss of -10,000 HU (default value for segmentation failure detected by the tool) outside the reference range. Therefore, the tool was considered a technical failure for the bone tool. The failure was attributed to the presence of spinal fusion hardware. Credit: ARRS/AJR
According to an accepted manuscript published in the American Journal of Roentgenology (AJR) by ARRS, certain reasons for AI tool failure related to technical factors can be largely prevented by proper acquisition and reconstruction protocols.
“The automated AI body composition tools demonstrated high technical appropriateness rates in a heterogeneous sample of external CT exams, supporting the tools’ generalizability and potential for widespread deployment,” concluded lead researcher B. Dustin Pooler, MD, of the University of Wisconsin School of Medicine and Public Health in Madison.
This AJR-accepted manuscript included 8,949 patients (mean age 55.5 years; 4,256 males, 4,693 females) who had undergone abdominal CT – performed at different facilities with different scanners from different manufacturers – and subsequently sent to the local PACS for clinical purposes were transferred. Using three independent automated AI tools to assess body composition via bone attenuation, muscle quantity and attenuation, and visceral and subcutaneous fat amounts, one axial series per examination was evaluated.
Ultimately, three fully automated AI tools measuring body composition (vertebral bones, body wall muscles, and abdominal visceral and subcutaneous fat) had technical adequacy rates of 97.8% to 99.1% on Pooler et al.’s sample of 11,699 external abdominal CT scans—performed at 777 different external institutions with 82 different scanner models from 6 different manufacturers.
Noting that the reasons for failure also include factors inherent in patients that are more difficult to control, “Explainability and an understanding of the reasons for failure can help build trust in AI tools and acceptance among radiologists and other physicians,” the authors of this AJR acknowledged added a manuscript.
For more information: B Dustin Pooler et al., Technical Adequacy of Fully Automated Artificial Intelligence Body Composition Tools: Assessment in a Heterogeneous Sample of External CT Examinations, American Journal of Roentgenology (2023). DOI: 10.2214/AJR.22.28745
Provided by the American Roentgen Ray Society
Citation: Technical Adequacy of Artificial Intelligence Body Composition Assessed in External CT (2023 February 23) Retrieved February 23, 2023 from https://medicalxpress.com/news/2023-02-technical-adequacy-artificial-intelligence- body.html
This document is protected by copyright. Except for fair trade for the purpose of private study or research, no part may be reproduced without written permission. The content is for informational purposes only.