A Comparison of Automated Segmentation and Manual Tracing of Magnetic Resonance Imaging to Quantify Lateral Ventricle Volumes

Abstract

Objective: Ventricular volume measurements have been proposed as a useful biomarker for several neurological diseases. The goal of this study was to compare the performance of 3 fully-automated tools, volBrain (http://volbrain.upv.es), ALVIN (Automatic Lateral Ventricle Delineation) (https://sites.google.com/site/mrilateralventricle/), and MRICloud (http://mricloud.org), with expert hand tracing to quantify lateral ventricle (LV) volume using magnetic resonance images. Materials and Methods: The sample comprised 24 healthy subjects (age: 25.1±5.7 years, all male). Volumes derived from each automated measurement were compared to hand tracing results performed by 2 specialists to assess the percent volume difference using the intraclass correlation coefficient (ICC), concordance correlation coefficient (CCC), Dice index value, and Bland-Altman analysis. Results: The ICC agreement of the Manual_1 and Manual_2 was very good (0.979), and there was no statistically significant difference (p>0.001). The volume difference of all methods was similar. The CCC with MRICloud and ALVIN was higher than that of volBrain. Bland-Altman plots indicated that the 3 automated methods demonstrated acceptable agreement. Conclusion: Compared with hand tracing, the LV volumes generated by MRICloud were more accurate than those of volBrain and ALVIN. LV volume values can provide valuable data related to the volumetric dependencies of the anatomical structures in various clinical conditions that can now be easily obtained using automated tools.