Performance of automatic Liver Volumetry for Selective Internal Radiotherapy

Ziel/Aim Liver volumetry plays a crucial role for therapy planning and dosimetry calculations in selective internal radiotherapy (SIRT). Several methods have been proposed in the literature and are available in commercial software. In our study, we aim at comparing the most common volumetry methods on a 101 patient-dataset with contrast-enhanced CT (ceCT) or MRI as pre-therapeutic imaging.

Methodik/Methods A 3D convolutional neural network (CNN) was trained to automatically segment the liver from T1-weighted DIXON MRIs on the water-only channel. 54 MRI and 47 ceCT datasets of 101 patients undergoing SIRT were evaluated in terms of the liver volume using the modified ellipsoid model [1], the BSA model [2], the interactive region growing algorithm of the ImFusion Suite [3] applied by a novice (1st year medical student), and a CNN (either the trained one for MRI or the one provided in the ImFusion Suite [3] for ceCT). The manual segmentations of an experienced physician served as ground truth.

Ergebnisse/Results The MRI CNN was trained to achieve a Dice Score of 92+-3 %. The BSA model resulted in a 25+-28 % error, while the modified ellipsoid model had an error of 30+-16 % and the interactive segmentation 5+-6 % (all three, n = 101). The MRI CNN yielded an error of 5+-6 % (n = 55) and the ceCT CNN 6+-21 % (n = 46). For the interactive segmentation and the CNNs the Dice score was 89+-7 %, 92+-4 % and 86+-15 %.

Schlussfolgerungen/Conclusions The liver volume estimations from BSA and ellipsoid models yield large deviations from ground truth liver volumetry thus resulting in large absorbed dose estimation errors. While the CNNs are fast and do not require human interaction, the interactive approach shows excellent results with a novice. The use of CNNs followed by refinement by interactive segmentation can take the best out of both approaches with minimal user-involvement, therefore increasing segmentation speed and reproducibility.

Publication History

Article published online:
08 April 2021

• Literatur/References

• 1 Vauthey JN, Abdalla EK, Doherty DA, Gertsch P, Fenstermacher MJ, Loyer EM, Lerut J, Materne R, Wang X, Encarnacion A, Herron D, Mathey C, Ferrari G, Charnsangavej C, Do KA, Denys A. Body surface area and body weight predict total liver volume in Western adults. Liver Transpl. 2002; Mar; 8 (03) 233-40. DOI: 10.1053/jlts.2002.31654. PMID: 11910568..
• 2 Muggli D, Müller MA, Karlo C, Fornaro J, Marincek B, Frauenfelder T. A simple method to approximate liver size on cross-sectional images using living liver models. Clinical Radiology 2009; Jul; 64 (07) 682-9.
• 3 ImFusion Suite, ImFusion GmbH, Munich, Germany, url: https://www.imfusion.com