Nuklearmedizin 2021; 60(02): 139
DOI: 10.1055/s-0041-1726799
WIS-Vortrag
Schilddrüse und Endokrinologie

Tracked 3D ultrasound and deep neural network-based thyroid segmentation improve accuracy and interobserver variability in thyroid volumetry

C Eilers
1   Technical University of Munich, Computer Aided Medical Procedures and Augmented Reality, Garching near Munich
,
M Krönke
2   Technical University of Munich, School of Medicine, Department of Nuclear Medicine, Munich
,
D Dimova
1   Technical University of Munich, Computer Aided Medical Procedures and Augmented Reality, Garching near Munich
,
M Köhler
1   Technical University of Munich, Computer Aided Medical Procedures and Augmented Reality, Garching near Munich
,
L Konstantinidou
1   Technical University of Munich, Computer Aided Medical Procedures and Augmented Reality, Garching near Munich
,
N Navab
1   Technical University of Munich, Computer Aided Medical Procedures and Augmented Reality, Garching near Munich
,
K Scheidhauer
2   Technical University of Munich, School of Medicine, Department of Nuclear Medicine, Munich
,
WA Weber
2   Technical University of Munich, School of Medicine, Department of Nuclear Medicine, Munich
,
J Nagarajah
3   Radboud University Medical Center, Department of Radiology and Nuclear Medicine, Nijmegen, the Netherlands
,
T Wendler
1   Technical University of Munich, Computer Aided Medical Procedures and Augmented Reality, Garching near Munich
› Institutsangaben
› Weitere Informationen
Auch verfügbar auf
 

Ziel/Aim Thyroid volumetry is crucial in diagnosis, treatment and monitoring of thyroid diseases. However, conventional thyroid volumetry with 2D ultrasound (US) is highly operator-dependent. This study compares 2D US and tracked 3D US with an automatic thyroid segmentation regarding inter- and intraobserver variability. Volume reference standard was MRI.

Methodik/Methods 21 healthy volunteers (24 - 50 a) were studied by 2D and 3D US as well as by MRI. Three physicians (MD1, 2 and 3) with different levels of experience (6, 4 and 1 a) performed three 2D US and three tracked 3D US scans on each volunteer. In the 2D scans the thyroid lobe volumes were calculated with the ellipsoid formula (correction factor: 0.48). A convolutional neural network (CNN) segmented the 3D thyroid lobes automatically. On MRI (T1 vibe sequence) the thyroid was manually segmented by an experienced MD.

Ergebnisse/Results MRI thyroid volumes ranged from 3.2 - 16.8ml (mean 7.98, SD 3.51). The interobserver variability comparing 2 MDs showed mean differences for 2D and 3D respectively of 0.95ml to 0.64ml (MD1 vs. 2), -2.33ml to -0.25ml (MD1 vs. 3) and -3.30ml to -0.90ml (MD2 vs. 3). Paired samples t-tests showed significant differences in all comparisons for 2D (p=.002, P<.001 and P<.001) but none for 3D (p=.55, p=.52 and p=.74). Intraobsever variability was similar for 2D and 3D US. Comparison of the mean US volumes and the MRI volumes by paired samples t-tests showed a significant difference for the 2D volumetry of MD1 (p=.001) and MD3 (p<.001), but none for any MDs for the 3D acquisitions (p(MD1)=.62, p(MD2)=.05, p(MD3)=.29).

Schlussfolgerungen/Conclusions Tracked 3D US combined with a CNN significantly reduces interobserver variability in thyroid volumetry and increases the accuracy of the measurements. Evaluations of thyroid morbidities are planned to further investigate the clinical relevance of this volumetric technique.



Publikationsverlauf

Artikel online veröffentlicht:
08. April 2021

© 2021. Thieme. All rights reserved.

Georg Thieme Verlag KG
Rüdigerstraße 14, 70469 Stuttgart, Germany