Toward a real time multi-tissue Adaptive Physics-Based Non-Rigid Registration framework for brain tumor resection

Abstract

This paper presents an adaptive non-rigid registration method for aligning pre-operative MRI with intra-operative MRI (iMRI) to compensate for brain deformation during brain tumor resection. This method extends a successful existing Physics-Based Non-Rigid Registration (PBNRR) technique implemented in ITKv4.5. The new method relies on a parallel adaptive heterogeneous biomechanical Finite Element (FE) model for tissue/tumor removal depicted in the iMRI. In contrast the existing PBNRR in ITK relies on homogeneous static FE model designed for brain shift only (i.e., it is not designed to handle brain tumor resection). As a result, the new method (1) accurately captures the intra-operative deformations associated with the tissue removal due to tumor resection and (2) reduces the end-to-end execution time to within the time constraints imposed by the neurosurgical procedure. The evaluation of the new method is based on 14 clinical cases with: (i) brain shift only (seven cases), (ii) partial tumor resection (two cases), and (iii) complete tumor resection (five cases). The new adaptive method can reduce the alignment error up to seven and five times compared to a rigid and ITK's PBNRR registration methods, respectively. On average, the alignment error of the new method is reduced by 9.23 and 5.63 mm compared to the alignment error from the rigid and PBNRR method implemented in ITK. Moreover, the total execution time for all the case studies is about 1 min or less in a Linux Dell workstation with 12 Intel Xeon 3.47 GHz CPU cores and 96 GB of RAM.

Keywords: ITK; biomechanical model; finite element method; non-rigid registration; real time; tumor resection.

Figure 1

The APBNRR framework. The red boxes represent the new contributions and the gray boxes the existing ITK modules. The red arrows show the execution order of the different modules. The APBNRR loop breaks when the desired number of iterations has reached. The output image (orange box) is the warped pre-op MRI when i = N_iter.

Figure 2

The correction of the warped pre-op segmented MRI. (A) Input warped pre-op segmented MRI. (B) Input intra-op mask MRI. (C) The (A) and (B) overlapped. (D) Output corrected warped pre-op segmented MRI at ith iteration.

Figure 3

Representation of the map between the corrected region of the warped pre-op segmented MRI (left) and the image deformation field DF_addi (right).

Figure 4

The Hausdorff Distance (HD) alignment error for the 14 clinical cases. The horizontal lines illustrate the average HD error of each method.

Figure 5

Qualitative evaluation results for 8 of the 14 case studies. Each row represents a single case. The left margin indicates the number and the type of each case. From left to right column: pre-op MRI, intra-op MRI, warped pre-op MRI (PBNRR), warped pre-op MRI (APBNRR), warped pre-op MRI (PBNRR) subtracted from intra-op MRI, warped pre-op MRI (APBNRR) subtracted from intra-op MRI. For the PTR and CTR cases, the cyan color delineates the tumor segmentation in the pre-op MRI. The BS cases do not require tumor segmentation.

Figure 6

Enhanced views of the intra-op MRI (case 13). The left figure demonstrates the brain prior to the extraction from the skull with background noise clearly visible. The right figure shows the brain following extraction from the skull. The background noise is mostly removed except the green demarcated region near the edges of the tumor cavity.

Figure 7

The PBNRR total execution time for the 14 clinical cases using 1, 4, 8, and 12 hardware cores.

Figure 8

The APBNRR total execution time for the 14 clinical cases using 1, 4, 8, and 12 hardware cores. The gray dashed line illustrates a threshold of 60 s.

Figure 9

The end-to-end speedup for the 14 clinical cases using 12 hardware cores.