P01.091 A robust method for rapidly simulating TTFields distributions within patient-specific computational head models. (19th September 2018)

• Record Type: Journal Article
• Authors: Urman, N; Levi, S; Frenkel, A; Naveh, A; Manzur, D; Hershkovich, H S; Wenger, C; Kirson, E; Bomzon, Z
• Is Part Of: Neuro-oncology. Volume 20(2018)Supplement 3
• Page Start: iii251
• Page End: iii251
• Publication Date: 2018-09-19
• Subjects: Brain Neoplasms -- Periodicals; Brain -- Tumors -- Periodicals; Brain -- Cancer -- Periodicals; Nervous system -- Cancer -- Periodicals; 616.99481
• DOI: 10.1093/neuonc/noy139.133
• Languages: English
• ISSNs: 1522-8517

Abstract

Abstract: Background: Tumor Treating Fields (TTFields) are delivered to the brain through two pairs of transducer arrays placed on the patient's scalp. TTFields distribution in the brain depends on the array's position, patient anatomy and the electric properties of the tissues and tumor. We aim to create realistic computational patient models in order to understand how field distribution influences disease progression. However, preparation of such models is a highly time consuming task. In addition, in clinical scenarios, MRI acquisition time is often reduced by increased slice spacing, limited field of view, or increased scan speed, leading to difficulties in creating automated models. Our method is designed to enable model creation under those restrictions. It does so by using a realistic head model of a healthy individual used as a deformable template with which the patient model is derived. Methods: A highly detailed healthy head model serves as a deformable template from which patient models are created. The first step is pre-processing, involving denoising and background noise reduction, as well as super-resolution algorithms when needed. To create the patient model, the tumor is first segmented manually and masked, leaving only healthy tissues in the MRI, which is then registered to the template space to yield the transformation from patient space to template space. The template is then deformed into the patient space using the inverse transformation, and the tumor is placed back creating a full patient model. Next, automatic identification of landmarks on the patient's head is used to position the transducer arrays on the head, which are then introduced into the model. Finally, boundary conditions are set, and field distribution is simulated using Finite Differences Time Domain (FDTD) method (Sim4Life V3.0, ZMT-Zurich). Results: We have simulated TTFields distribution of 317 patients treated with TTFields as part of the EF-14 trial. Our method is optimized for accurate contouring of tissues highly influencing the distribution of electric field (Scalp, skull, CSF, ventricles), and it is robust, having the ability to give sufficient results even when MRI data quality is low. Thus enabling a study correlating the spatial distribution of TTFields and patient outcome. Conclusion: Our process for rapidly creating patient models presents a breakthrough that enables the first study in which the spatial distribution of therapeutic electric fields correlates with patient outcome. In the future this method can be used for clinical studies investigating other clinical indications of large datasets.