Head model dataset for mixed reality navigation in neurosurgical interventions for intracranial lesions

Abstract

Mixed reality navigation (MRN) technology is emerging as an increasingly significant and interesting topic in neurosurgery. MRN enables neurosurgeons to "see through" the head with an interactive, hybrid visualization environment that merges virtual- and physical-world elements. Offering immersive, intuitive, and reliable guidance for preoperative and intraoperative intervention of intracranial lesions, MRN showcases its potential as an economically efficient and user-friendly alternative to standard neuronavigation systems. However, the clinical research and development of MRN systems present challenges: recruiting a sufficient number of patients within a limited timeframe is difficult, and acquiring low-cost, commercially available, medically significant head phantoms is equally challenging. To accelerate the development of novel MRN systems and surmount these obstacles, the study presents a dataset designed for MRN system development and testing in neurosurgery. It includes CT and MRI data from 19 patients with intracranial lesions and derived 3D models of anatomical structures and validation references. The models are available in Wavefront object (OBJ) and Stereolithography (STL) formats, supporting the creation and assessment of neurosurgical MRN applications.

Fig. 1

Practical workflow to produce the data in this study. Based on the enrollment criteria, CT/MRI data from 44 cases were collected in DICOM format. Undergoing a structured screening, data from 19 cases were chosen for further processing. On the one hand, the imaging data were reformatted, anonymized, segmented, and 3D reconstructed to generate holograms for visualization using MRN. On the other hand, the skin surface on each patient’s head was extracted and reconstructed from the data, then optimized for low-cost 3D printing and incorporated with validation reference objects. 3D = three-dimensional; CT = computed tomography; MRI = magnetic resonance imaging; MRB = medical reality bundle; MRN = mixed reality navigation; OBJ = object; STL = stereolithography.

Fig. 2

An illustration of the process of generating holograms. Subfigure (A) demonstrates the co-registration of a patient’s multimodal sequences into a unified coordinate system (indicated by the red dashed crosshairs), which is defined by the highest resolution reference image (RI). Following image fusion (B), synchronous observation is permitted, and segmentation is performed within the unified coordinate system (C). Subsequently, clusters of segmented voxels are transformed into a 3D surface model, i.e., holograms, which can be observed from any angle, not limited to the given imaging planes (D).

Fig. 3

An overview of the validation reference objects principle. Subfigures (A–C) illustrate the marker-based comparison. In the reference image (RI), centroids (green crosses) of the markers (red spheres) are automatically extracted within the image coordinate system (blue axes) and serve as the ground truth (A), while the physical head phantom is designed to incorporate the markers (blue spheres) (B). After registering the virtual content to the phantom using the MRN system, the user can capture the coordinates of the perceived physical points (blue spheres) in virtual space, allowing for the measurement of their deviation from the ground truth. Subfigures (D–F) demonstrate the comparison based on positioning lines. The hologram of scalp quadrants (cyan) is created using orthogonal reference planes and the segmented skin surface from the RI (D), while the physical head phantom integrates the laser positioning line models (red lines) (E). Once the virtual content is registered to the head phantom with the MRN system, users can observe the mismatch between the scalp quadrant and the physical model of positioning lines, providing an intuitive impression of the registration quality.

Fig. 4

The structure of the dataset (left), an example anonymized DICOM archive file (red box), and an example MRB file (right, blue box). The forward slash “/“ represents a directory.