A sparse intraoperative data-driven biomechanical model to compensate for brain shift during neuronavigation

Abstract

Background and purpose: Intraoperative brain deformation is an important factor compromising the accuracy of image-guided neurosurgery. The purpose of this study was to elucidate the role of a model-updated image in the compensation of intraoperative brain shift.

Materials and methods: An FE linear elastic model was built and evaluated in 11 patients with craniotomies. To build this model, we provided a novel model-guided segmentation algorithm. After craniotomy, the sparse intraoperative data (the deformed cortical surface) were tracked by a 3D LRS. The surface deformation, calculated by an extended RPM algorithm, was applied on the FE model as a boundary condition to estimate the entire brain shift. The compensation accuracy of this model was validated by the real-time image data of brain deformation acquired by intraoperative MR imaging.

Results: The prediction error of this model ranged from 1.29 to 1.91 mm (mean, 1.62 ± 0.22 mm), and the compensation accuracy ranged from 62.8% to 81.4% (mean, 69.2 ± 5.3%). The compensation accuracy on the displacement of subcortical structures was higher than that of deep structures (71.3 ± 6.1%:66.8 ± 5.0%, P < .01). In addition, the compensation accuracy in the group with a horizontal bone window was higher than that in the group with a nonhorizontal bone window (72.0 ± 5.3%:65.7 ± 2.9%, P < .05).

Conclusions: Combined with our novel model-guided segmentation and extended RPM algorithms, this sparse data-driven biomechanical model is expected to be a reliable, efficient, and convenient approach for compensation of intraoperative brain shift in image-guided surgery.

Fig 1.

A and B, Model-guided segmentation. The model (green) is aligned with the preoperative MR imaging to get a primitive result with a contour (bright gray) very close to the real contour (dark gray). C, Coarse mesh. D, Multiresolution mesh.

Fig 2.

A, Surface scanning by LRS (left) after registration with the Excelim-04 system (middle) and PoleStar N20 iMRI system (right). B, The deformed surface (right side) acquired by LRS scanning. C, Transformation from the LRS space to the image space. D, Registration of the initial surface with the deforming surface.

Fig 3.

A, The predictive deformation of mesh. The red mesh represents the preoperative surface, the blue mesh represents the deformed surface, and the yellow arrow indicates the direction of gravity. B, 3D visualization of the deformation field by the ray casting method (above) and the final warped MR images (below). The magnitude of the deformation increases as the color changes from dark red to bright red; the blue arrows near the frontal lobe indicate the direction of the deformation.