171 F.2 (a) Red points indicate examples of identified and collected cortical surface vessel points. via optically tracked stylus, displayed on the preoperative MR image recorded in the patient room in the initial phase of the resection procedure after craniotomy and opening of the dura. Yellow points indicate examples of the surface features collected after resection shown on preoperative MRI. c) Green points indicate those collected via an optically tracked stylus, demonstrating here the spatial distribution of the collected points, as well as their relationship to vascular structures, shown in the cortical surface view.
Specific Aims
The goal of Objective 1 is to develop a generalized validation framework that allows assessment of the accuracy of model-based brain shift correction strategies. The aim of Goal 2 is to develop and validate a patient-specific biomechanical model-based brain shift correction based on deformation atlas.
Dissertation Overview
Table V.6 shows that the brain shift correction performance of the FEM model-based approach is consistent, i.e. especially in Figure F.5 on the left, noting that the model predicted tissue swelling in the craniotomy area.
Background and Significance
Deep Brain Stimulation
Medical Image Guidance in DBS
Brain Shift in DBS Burr Hole Surgery
Brain Shift Compensation Strategies in DBS
- MER-Assisted Awake Surgery
- iMR-Guided Asleep Surgery
- Other Intraoperative Imaging-Based Approaches
- Biomechanical Model-Based Approaches
Summary
Method Overview
Overall Study Design
A specific focus of this Dissertation is the use of medical image data for surgical implantation of the DBS electrodes. Brain shift leads misalignment between preoperative condition of the patient and the intraoperative anatomy caused by soft tissue deformation due to burr hole and dural opening [29].
An iMR-Leveraged Validation Framework
- AMIGO Suite at BWH
- iMR-Guided DBS Surgery at UCSF
- Validation Framework with iMR
The iMR data used in this thesis, specifically for Chapter IV, were collected during brain tumor resection operations in the AMIGO Suite; details of the patients (including tumor types and grades) can be found in Table IV.2. To extend the gray block shown in Figure III.1 with the ultimate goal of achieving model accuracy assessment (grey checkerboard block in Figure III.2), the developed generalized validation framework is shown in Figure III.2.
Biomechanical Model
- Deformation atlas – Boundary Condition Assignment
- Biphasic Biomechanical Model
- Inverse Problem Approach
Gravity (assuming the patient is supine) is also shown in Figure III.4 in blue vector [75]. This new approach constructs the M-matrix in equation III.5 at the measurement point instead of at the nearest mesh node, as in our previous work:
Bioelectric Model
A bioelectrical model can also provide assessment of the volume of tissue activation (VTA), which can then be used to predict (i) tractography, aided by additional information such as diffusion tensor imaging (DTI), thereby providing insight into the extent of neural responses via the stimulations of the axons [91]; and/or (ii) merge with the patient image data to investigate the overlap between the activation zone and anatomical structures for surgical planning and target selection [91]. One additional consideration of the conduction properties of the computational medium is the encapsulation layer around the electrode [91, 109].
VTA and Tractography
Here, the differences between intraoperative and model-predicted positions represent the residual error in the model-based approach. A flowchart of the study: statement of various switching considerations (model and ANTs predictions in black block) within the bioelectrical model (blue block) for VTA estimations in HCP 1021 template (gray block), for subsequent examination of tractography (orange block) ).
An iMR-Leveraged Generalized Validation Framework
Summary and Contributions
This chapter presents a generalized validation framework that utilizes iMR imaging data for gold-standard intraoperative brain displacement measurements against which predicted biomechanical models are compared. Miga, “Validation of model-based brain displacement correction in neurosurgery via intraoperative magnetic resonance imaging: Preliminary results,” Proceedings of SPIE, 2017.
Abstract
Miga, “Retrospective study comparing model-based deformity correction with intraoperative magnetic resonance imaging for image-guided neurosurgery,” Journal of Medical Imaging , vol. We performed a retrospective study of n=9 tumor resection cases, comparing iMR measurements with intraoperative brain displacement compensation predicted by our model-based strategy driven by sparse intraoperative cortical surface data.
Introduction
There have been several research efforts to develop predictive biomechanical modeling driven by sparse intraoperative data to compensate for brain motion. Many studies have used physical or digital phantoms as a preliminary validation step to assess the accuracy of their proposed brain motion correction algorithms [ 64 - 66 ].
Methods
- Preoperative Image Information and Segmentation
- Preoperative Computational Biomechanical Model Construction
- Deformation Atlas Generation
- Homologous Surface Point Selection and Inverse Problem
- Subsurface Shift Measurement, Prediction and
- Point Selection—Measurement of Uncertainty and Propagation of Error
Details of each patient's preoperative and intraoperative MRI can be found in Table IV.3. An example of the distribution of measured subsurface displacements around the tumor is shown in Figure IV.4(c) and (d) in red vectors.
Results
- Overall View: Analysis of Nine Cases
- Model Overall Performance Evaluation
- Point Selection Uncertainty and Propagation of Error
The result of the area correction, quantified by the residual error in equation IV.4, and the percent correction defined in equation IV.5, is shown in Table IV.5. The target registration error or residual error is described in equation IV.4 and the percent correction defined in equation.
Discussion
The result (i.e. residual distance) of the model correction shows a relative indifference to the small area selection uncertainty shown in Table IV.9. When examining the subsurface deformation shown in Figure IV.8(a) - (c), the direction of the subsurface deformation shows a strong tendency to collapse into the tumor.
Conclusion
A critical component of the developed model-based approach to DBS is the ability to incorporate subsurface data, e.g. An example of the obtained data is shown in Figure VI.2, where the displacement of the brain due to the punch-hole procedure is demonstrated both at the surface and at the subsurface (shown by the corresponding intersections in the lateral ventricle).
Accounting for Brain Shift in DBS Surgery Leveraging Sparse Data via
Summary and Contributions
In particular, a similar non-negligible shift magnitude at the target structure in the deep brain, or bulk tissue motion near the target area (model 1.2 mm and ANTs 1.4 mm), is predicted by both the model-based approach and the non-rigid image registration method. to illustrate the potential clinical need and utility of the developed biomechanical modeling method. Additional material detailing additional co-authored studies using the adaptation of the inverse problem approach with subsurface input data is also provided in the final section of this chapter, reprinted with permission from.
Abstract
Introduction
An example of the obtained data is shown in Figure V.1, where significant asymmetric shift is readily observable. An example of the collected points (red points) after craniotomy and dura opening with the preoperative MR image registered to the patient.
Methods
- Data
- Biomechanical Model-based Deformation Atlas
- Inverse Problem Approach
- Model Performance Assessment
- Estimation of Brain Shift at Surgical Target Region
- Comparison to Nonrigid Image Registration
- Shift Correction Performance on Parenchymal Targets
- Shift Estimation at Surgical Target Region
Discussion
Furthermore, given the imperfect nature of non-rigid registration and the aforementioned potential intra-operator error, the residual error of ANTs with respect to the subsurface points suggests the reliability and validity of the point designation. Finally, although the aim of this study is to investigate the feasibility and establish potential accuracy metrics for a model-based approach to cope with brain shifts in DBS under an experimental design compatible with the data sparsity expected in surgery, to the clinical translatability of Due to the method, it is important to assess and consider the potential challenges associated with the deployability and usability of the proposed approach in the OR, especially with regard to obtaining scarce intraoperative measurements that are needed to address the inverse problem through OR-compatible and -friendly means, as this is likely to be the rate-limiting step of the proposed approach.
Conclusion
The use of iCT in DBS pinhole surgery has been demonstrated by groups such as Burchiel et al. For the model-based approach proposed here, ICT is likely to provide intraoperative surface deformation information to drive the inverse problem. however, again the poor soft tissue contrast of ICT would limit the finding of relevant features. ii) Another alternative route is to use subsurface data via US (transcranial or borehole).
Supplemental Material
- Summary and Contributions
- Method Overview
- Results
- Discussions and Conclusions
An example of image data in this study: demonstration of the impact of brain shift during DBS burr surgery on the surface as well as subsurface (corresponding cross hair at the lateral ventricle). This is a preliminary multi-physics framework that examines the feasibility of the aforementioned framework as well as the impact of brain shifting on tractography.
Predictive Multi-Physics Modeling Framework in DBS
Summary and Contributions
Miga, "An Integrated Multi-Physics Finite Element Modeling Framework for Deep Brain Stimulation: Preliminary Study on the Impact of Brain Shifting on Neuronal Pathways," Medical Imaging Computing and Computer-Assisted Intervention, Lecture Notes in Computer Science (2019). Miga, "Predictive multi-physics modeling framework in deep brain stimulation: investigating the impact of brain shifting on neural pathways."
Abstract
This coupled biomechanical and bioelectrical modeling framework offers a comprehensive intraoperative platform with multiphysics predictive capability and clinical applicability that requires only sparse intraoperative data. Conclusions: The predictive multi-physics framework presented herein may enable direct visualization, targeting, and prediction of neural pathway recruitment.
Introduction
However, in most studies of the model-based approach to investigating brain displacement compensation, model placement is often done after a significant resection has occurred. Qualitative comparisons are made between the preoperative MR image of the patient and the image of the deformed model due to swelling and collapse in Figure F.7.
Methods
- Data
- Accounting for Brain Shift via Biomechanical Model
- Computing Potential Distribution via Bioelectric Model
- Estimating Volume of Tissue Activation and Tractography
- Assessing Shift Impact on Neural Pathway
Results
- Estimated Shift at Deep Brain Target
- VTA Estimation
- Tractography Prediction
- Volumetric Overlap (Jaccard Index) with Shift Estimation
- Connectivity Profile
Finally, the volumetric overlap is shown with the corresponding displacement estimate in Table VI.1, shown in Figure VI.8. A linear regression was performed with Pearson CC (in Figure VI.8), with model prediction and ANT prediction considered separately (left and middle) as well as together (right).
Discussion
Furthermore, the estimated threshold at which the target activation drops to 0%, or simply the reciprocal of the slope of the linear regression (as shown in Figure VI.8) is and 2.61 mm for active contacts 0, 2 and 3, respectively. Another potential limitation of the study is the lack of specific diffusion-weighted imaging (DWI) data.
Conclusion
The iMR validation effort in Chapters IV and V represents progress in understanding the brain drift correction capabilities of the model-based approach. The objective of Appendix B is to illustrate the application of Galerkin's weighted residual method to a two-phase model.
Future Directions and Conclusion
Prospective Deployment and Validation of the Modeling Method
- Longitudinal Image-Update Capability of the Model-Based Approach
- Initial Clinical Experience: Model-Based Shift-Accounted Image-Updating
Here in Figure VII.4(b), it is worth noting as a nearby comparator the Medtronic StealthStation (Medtronic Inc., Minneapolis, MN). Shown in Figure VII.6(a), the surgical team evaluated the performance of the model by comparing surface and subsurface structures (such as the tumor cavity) with the system's displacement-computed MR display in Figure VII.6(b) with the Medtronic display of rigid registration in figure VII.6(c).
Incorporation of Intraoperative Imaging for
However, the US probe in Chapter V.8 (a neuro-cranial probe) is likely to have greater imaging capability (eg volume) than the pinhole probe here, future studies are needed to examine the feasibility (ie, lateral resolution and signal-to-noise ratio, or SNR) of using the pinhole probe to capture the shape of the ventricle.
Augmentation of the Predictive Framework via Machine Learning
To unify different mesh shapes as well as increase the amount of training data, a dense uniform grid could be described in 'atlas' patient image space, and displacement solutions (now including all patient data transformed into this common space) could be interpolated on these grid points. Once a network is trained with this setup, when a new patient is presented, as long as transformation is achieved (e.g. via ANTs) between that patient's pMR and 'atlas' MR, and features such as gravity are correctly described as mentioned above. , efficient shift predictions could be possible with the trained network driven by input from intraoperative surface displacement measurement.
Concluding Remarks
The purpose of Appendix C is to illustrate the application of the Galerkin weighted residual method to the Poisson equation. With model prediction, the preoperative MR image can be updated to reflect patient anatomy intraoperatively due to swelling at the initial stage of the procedure, i.e.
Biphasic Model Derivations
Application of the Galerkin Weighted Residual Method to the Biphasic Model
Application of the Galerkin Weighted Residual Method to the Poisson Equation
Patient iUS Images Superimposed with Preoperative MR
Initial Development Toward a Multi-Physics Modeling Framework
Summary and Contributions
Miga, "An Integrated Multiphysics Finite Element Modeling Framework for Deep Brain Stimulation: Preliminary Study on the Impact of Brain Displacement on Neuronal Pathways", Medical Computer Imaging and Computer Assisted Intervention - MICCAI 2019, Lecture Notes in Computer Science, vol 1176.
Abstract
Specifically, the biomechanical model was used to predict brain change via an inverse problem approach, which was driven by sparse data derived from interventional magnetic resonance (iMR) imaging data. Therefore, consideration of brain replacement in DBS bursal surgery is desired to optimize the result.
Introduction
Comprehensive efforts for brain displacement compensation in DBS surgery via model-based approach are quite limited, especially in terms of validation studies using clinical patient data and possible intraoperative implementation. In the work presented here, we examine the feasibility of this integrated multi-physics framework of patient-specific biomechanical and bioelectrical models and examine the differences in VTA and neuronal pathway recruitment with and without considering brain shift.
Methodology
- Data
- Biomechanical Model-based Approach for Brain Shift Estimation
- Bioelectric Model for Volume of Tissue Activation Estimation
- Integrated Framework—Impact of Shift on Neuronal Recruitment
With the lead insertion path visible in the iMR image shown in Fig. E.1(b), the path of the lead leads was determined by identifying two points along the aforementioned path. With the specified model-estimated VTA in the intraoperative configuration and the obtained model-predicted brain displacement field, the active nodes forming the VTA could be mapped to the preoperative space by inversion of the model displacement field.
Result
- Brain Shift Compensation Performance
- VTA Estimation
- Neuronal Pathway Recruitment with and without Shift Consideration
The bioelectrical model with reconstructed electrode wires is shown in Figure E.2(a) and (b), where the model reports the predicted asymmetric shift. The estimated VTAs are shown in yellow and overlaid on the updated model MR in Figure E.2(c)-(d) for case 1 and case 2, respectively.
Discussion and Conclusions
The blue crosshairs indicate in particular that the model was able to depict the border of the ventricle better. The capture of these modes of deformation in the course of surgery representing a longitudinal use of the model-based approach is somewhat novel, i.e.
Longitudinal Application of Model-Based Brain Shift Correction: A Case Study
Summary and Contributions
This work represents a clinical case study where the model-based brain shift correction strategy was employed at two different time points of surgery, ∼2 hours apart, to address swelling-induced and subsidence-induced brain shift, driven by surface data obtained by optically tracked stylus . Qualitative and quantitative assessments illustrate the feasibility of using a model-based approach in the OR, especially with regard to longitudinal application, i.e.
Abstract
Miga, “A comprehensive model-supported approach to brain motion correction in image-guided neurosurgery: a case study of brain swelling and subsequent drooping after craniotomy,” Proceedings of SPIE, 2019. Intraoperative tissue swelling and drooping were captured with an optical tracking pen identifying features of cortical superficial vessels (n = 9) and a model-based correction was performed for these two different types of brain drift at different stages of the process.
Introduction
Throughout the operation, we estimate that the cortical surface experienced a deformation trajectory with an absolute path length of approximately mm reflecting swelling followed by sagging. Thus, in addition to the longitudinal aspect of model validation, this work represents an in vivo case study that examines the performance of the model when challenged with a scenario.
Methodology
- Overview
- Data Acquisition
- Deformation-Atlas Model-Based Brain Shift Correction
- Qualitative and Quantitative Assessment of Model
An example of the collected points (yellow points) at a later stage of the procedure after resection mapped to the preoperative MR image is shown in Figure F.2(b), showing the effect of brain tissue subsidence, as the collected points below the brain surface of the preoperative MR image. To measure displacement required to drive the inverse problem approach, the vessel points collected during the initial stage of the procedure (after the craniotomy and dura opening) were projected onto the brain mesh generated from the preoperative MR image .
Result
Specifically, for the collected surface points, the intraoperative measurements shown in Figure .4 are compared to the model prediction and the difference is defined as the residual error of the model. In addition, iMR shows that there was a shape change in the ventricle compared to the patient's preoperative anatomy, which the deformed MR model was able to recover to some extent.