Creating Patient-Specific Computational Head Models for the Study of Tissue-Electric Field Interactions Using Deformable Templates

Abstract

Understanding how electric and magnetic fields interact with the human head is important for a variety of applications such as determining the safety of MRI devices and sequences, studying trans-cranial direct current stimulation and Tumor Treating Fields (TTFields) treatment planning for Glioblastoma Multiforme (GBM) patients. These topics can be studied by creating realistic computational head phantoms and numerically calculating the electro-magnetic field distribution within these models. However, the preparation of such computational phantoms involves accurate segmentation of MRI images, a task that cannot be performed automatically. Therefore, creating the phantoms is time-consuming. This is particularly true when segmenting the MRIs of patients with head tumors, because existing segmentation algorithms are unable to account for the presence of the abnormal tissue and therefore often fail to identify the tumor and surrounding tissue in a satisfactory manner. To overcome this limitation, and enable the creation of realistic computational phantoms with head tumors, we have developed a method in which a realistic head model of a healthy individual is used as a deformable template with which the patient model is derived. A prerequisite for our method is the creation of a highly detailed healthy head model which serves as a deformable template from which patient models can be created. The template can be created using existing procedures. When creating patient models with our method, the first step is to manually segment the tumor from the patient’s MRI. The region of the tumor in the patient MRI is masked, and non-rigid registration algorithms are used to register the remaining regions of the patient head on to a 3D discrete image representing the deformable template. This process yields a non-rigid transformation that maps the healthy portion of the patient's head in to the template space, as well as the inverse transformation that maps the template in to the patient space. The inverse transformation is applied to the 3D deformable template to yield an approximation of the patient head in the absence of a tumor. Finally, the tumor is planted back into the deformed template to yield the full patient model. Comparison of the models to patient MRIs of varying quality and resolution reveals good representation of the patient anatomy within the model, even when the patient MRIs have low resolution, signal to noise, or contrast. Our results show that the creation of patient-specific head models using deformable templates is a robust and rapid procedure that requires minimal user input. This process is not only useful for studies on TTFields, but also for other applications that require modelling the interaction of electromagnetic energy with the human head.