The study compared the performance of an internally developed tool and the publicly available TotalSegmentator tool in segmenting muscle, subcutaneous fat, and visceral fat from CT scans. The researchers used the publicly available SAROS dataset, which contains 900 CT series from 882 patients, to evaluate the tools.
For subcutaneous fat segmentation, the internal tool achieved a 3% higher Dice score compared to TotalSegmentator (83.8 vs. 80.8). For muscle segmentation, the internal tool showed a 5% improvement in Dice score (87.6 vs. 83.2). The results were statistically significant (p < 0.01).
Due to the lack of ground truth segmentations for visceral fat in the SAROS dataset, the researchers used Cohen's Kappa to assess the agreement between the two tools. The Kappa score of 0.856 indicated a near-perfect agreement between the tools in segmenting visceral fat.
The internal tool also demonstrated strong correlations with the ground truth annotations for muscle volume (R^2 = 0.99), muscle attenuation (R^2 = 0.93), and subcutaneous fat volume (R^2 = 0.99). The correlation for subcutaneous fat attenuation was moderate (R^2 = 0.45).
The Bland-Altman analysis showed that the internal tool had a significantly lower bias in muscle volume estimation compared to TotalSegmentator. For subcutaneous fat volume, the internal tool had a slightly higher positive bias.
The study highlights the potential of the internally developed tool in advancing the accuracy of body composition analysis, which is crucial for various medical applications, such as disease characterization, surgical planning, and personalized risk assessment.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Benjamin Hou... at arxiv.org 04-16-2024
https://arxiv.org/pdf/2401.05294.pdfDeeper Inquiries