BrainStereo: clinical application and efficiency evaluation of an open-source stereotactic planning tool

Abstract

Objective: To develop a flexible and open-source stereotactic surgical planning toolkit, validated through clinical data to assess its performance in frame registration and stereotactic neurosurgical planning.

Methods: BrainStereo was developed based on the Leksell stereotactic frame principles and the 3D Slicer platform. It features an interactive interface for frame registration based on the custom-designed Layerwise Max Intensity Tracking (LMIT) algorithm, automated target/entry point calculation, and real-time 3D visualization. A retrospective analysis of stereotactic CT data from two hospitals was conducted, comparing BrainStereo with standard planning software to evaluate accuracy and efficiency.

Results: BrainStereo was developed as a comprehensive toolkit integrating frame registration, target and entry point computation, and dynamic 3D visualization. A total of 86 CT datasets from two hospitals were included. The root mean square error (RMSE) for frame registration was 0.56 ± 0.23 mm. Computation time for BrainStereo was 5.54 ± 1.16 min, significantly longer than the standard toolkit (4.75 ± 0.83 min, 95% CI: 4.57-4.92 min, p = 0.001), but showed a steeper learning curve. The mean Euclidean distance between target points from both toolkits was 0.82 ± 0.21 mm (95% CI: 0.74-0.90 mm), with no significant differences along the X, Y, and Z axes. Entry point deviations were 0.47° ± 0.37° (p = 0.07 for arc and p = 0.06 for ring). Bland-Altman analysis confirmed strong agreement, supporting BrainStereo's reliability for stereotactic neurosurgical planning.

Conclusions: BrainStereo is an open-source stereotactic planning tool that provides neurosurgeons and researchers with a flexible alternative to proprietary software. Integrated within 3D Slicer, it allows for adjustable parameters and modular functionality, addressing some of the limitations commonly associated with commercial solutions, such as hardware restrictions and limited adaptability. By offering open-source access, BrainStereo fosters transparency, collaboration, and broader accessibility, potentially advancing the field of stereotactic neurosurgery.

Keywords: Leksell frame; Neurosurgery; Open-source toolkit; Stereotactic.

Fig. 1

Structure of the Leksell stereotactic frame and illustration of arc and ring angles. (A) Anterior view of the Leksell stereotactic frame. (B) Demonstration of the arc angle on the Leksell frame model. (C) Lateral (left) view of the Leksell frame. (D) Demonstration of the ring angle on the Leksell frame model

Fig. 2

The RAS coordinate System and Leksell Frame model. (A) Comparison between the RAS and Leksell Cartesian coordinate systems, which share a common origin. The RAS system is shown in black, and the Leksell system in white. The yellow arrow indicates the axis oriented perpendicularly to the viewing plane, pointing from posterior to anterior. This axis corresponds to the A-axis in the RAS system and the Y-axis in the Leksell coordinate system. while the directions of the other two axes are reversed. (B) Dimensions of the Leksell stereotactic frame, including the overall width and the characteristic N-shaped fiducial board. (C-E) Orthogonal views of the Leksell frame: anterior (front), left lateral, and superior (top) perspectives

Fig. 3

Vertex identification based on the LMIT algorithm. (A) Four roughly selected points (green) are placed on an arbitrary slice. (B) The algorithm then performs upward layer-by-layer tracking, automatically marking the path in red until the vertex is identified. (C) To simulate user variability, five sets of initial points (green) were assumed to reflect possible manual placement deviations. (D) The software expands a 10-pixel region around each point and automatically locates the voxel with the highest CT value, correcting each point to that location (blue)

Fig. 4

Comparison of calculation time between two methods. (A) Scatter plot of calculation time and quadratic regression analysis for BrainStereo. (B) Scatter plot of calculation time and quadratic regression analysis for the standard software. (C) Box plot comparing the calculation times of BrainStereo and the standard software, showing that BrainStereo takes longer on average

Fig. 5

Frame registration accuracy in BrainStereo. (A) Frequency distribution of the root mean square error (RMSE) in frame registration using BrainStereo. (B) Scatter plot of RMSE across different cases, with a fitted trend line shows that the RMSE remains relatively stable across different cases

Fig. 6

Target and entry accuracy between two ways. (A) Box plot comparing the absolute deviations of BrainStereo and standard software along the X,Y and Z axis. (B) Histogram of the Euclidean distances between the calculated target points of BrainStereo and those of the standard software. (C) Results of two software based on arc. (D) Results of two software based on ring