“Sulcus Sink”: A Compact Binary and Semi-Automated Inverse Dijkstra-based System for Describing Sulcal Trajectories

Rudolph Pienaar, Christian Hasselgrove, Kiho Im, David N. Kennedy, P. Ellen Grant, Denise Boriel, Lena Tang, Nikos MakrisbioRxiv 2020 · DOI

Abstract

We present a system that uses a compact binary representation to describe and trace sulci on a reconstructed human cortical surface, based on a set of human-generated targets. The inputs to the system were manually created on a training set of 20 normal subjects (11 females, 9 males) with ages 22–40 years. T1-weighted MPRAGE images were collected and reconstructed with FreeSurfer, and 10 sulci on each hemisphere were traced by an expert human operator. The system determines a compact binary feature vector for each sulcus using a binary parametrized function of surface-geometry variables (e.g., mean curvature, sulcal depth, edge length). This function was optimized using a Dijkstra-based graph theory formulation. Individual subjects showed 91% average overlap within 4mm of the human target, while generalized group results achieved 80% overlap. This parametrized binary weight description provides a mechanism to succinctly describe individual and groups of human sulci.

Introduction

The human cortex is a highly folded surface, and the patterns of these folds—the sulci—are critical for brain mapping, surgical planning, and studying neurodevelopment. However, identifying and tracing these sulci is a major bottleneck in neuroscience. Manual tracing by experts is the "gold standard" but is incredibly time-consuming and subjective. Conversely, fully automated methods often struggle with the high variability of human brains or produce results that lack anatomical logic. The "Sulcus Sink" system addresses this by capturing expert intuition in a compact, mathematical "word," providing a standardized way to describe the logic used to navigate brain geometry.

Methodology

The researchers developed a methodology that treats the reconstructed 3D mesh of the brain as a graph where vertices are nodes and mesh edges are paths. The core of the system is a cost function based on nine geometric features, such as mean curvature, sulcal depth, and edge length.

Sulcal trajectories traced on the reconstructed cortical surface of a human brain.

To optimize this function, the system uses Inverse Dijkstra Optimization. While standard Dijkstra’s algorithm finds the "cheapest" path between two points given set costs, this system searches through 512 possible 9-bit binary weight combinations ($2^9$) to find the "recipe" of geometric features that makes the computer's path most closely match a human-traced ground truth. To handle "bumps" or interrupting ridges (gyri) that sometimes cross a sulcus, a non-linear penalty matrix toggles weights when curvature signs change. After individual optimization, weights were averaged to create "template binary words" for specific sulci.

Results and Key Insights

The study utilized a training set of 20 healthy subjects, with 10 specific sulci on each hemisphere (400 total trajectories) manually traced as ground truth.

Visualization of the Dijkstra-based path optimization showing the mesh (left) and the resulting cost surface (right).

Performance Metrics

  • Individual Accuracy: When optimized for a specific person, the system achieved a 91% average overlap (within 4mm) with the human expert's target.
  • Template Performance: Using a generalized "template word" for all subjects resulted in an 80% average overlap.
  • Expert Validation: In blinded tests, an independent anatomist rated most computer-generated paths as "largely acceptable" or "completely acceptable."

Anatomical Logic as a Binary Word

A striking insight is that complex anatomical trajectories can be described by a tiny 9-bit vector, suggesting that human experts follow a consistent, relatively simple geometric logic. By analyzing weight stability, the authors identified which geometric cues (like sulcal depth) are universal markers versus those that are more variable (like curvature). Interestingly, the system's template sometimes outperformed original human traces in expert evaluations, successfully generalizing the "true" path and smoothing out human inconsistencies.

Discussion and Future Work

One notable challenge was the cingulate sulcus, where performance was lower. This suggests that for very long sulci, the geometric logic may change mid-path, indicating they should be modeled as multiple joined segments.

While the 9-bit binary vector is compact, it is also coarse. Future work will explore larger bit-depths or additional parameters like Gaussian curvature to improve precision. Currently, the system is semi-automated, requiring human-defined start and end points; future iterations aim to automate these anchor points and explore more sophisticated clustering methods for template generation.

Summary updated

Abstract

Abstract We present a description of a system that uses a compact binary representation to describe and trace sulci on a reconstructed human cortical surface, based on a set of human-generated targets. The inputs to the system were manually created on a training set of 20 normal subjects (11 females, 9 males) with ages 22 – 40 years. T1 weighted MPRAGE images were collected on a Siemens 3T Trio scanner, with TR/TE = 2530/3.3, matrix = 256×256, FOV = 256mm, slice thickness 1.33mm. The resultant images were reconstructed with Freesurfer, and 10 sulci on each hemisphere were traced by an expert human operator and independently assessed for accuracy. Presented with these input trajectories in its training phase, the system attempted to determine a compact binary feature vector of each sulcus on each subject using as descriptor a binary parametrized function of several surface-geometry variables (such as mean curvature, sulcal depth, edge length, etc.). This function was optimized in a supervised learning fashion using a Dijkstra-based graph theory formulation, in which the binary weights were used to define graph edge costs. In the setup phase, the system was presented with sulcal trajectories already defined on surfaces, and then adjusted its parametrized weights in a binary fashion to minimize differences between the training input path and its Dijkstra-generated output path. Once the setup phase was complete and sulci had been described in a per-sulcus, per-subject manner, we generalized the per-sulcus description across all the subjects to construct a template binary word for each specific sulcus. The performance of the system for each subject and each sulcus, and for each template sulcus group was measured against the original human reference in both a quantitative and qualitative manner. Individual subjects generally showed very good optimization to their manually traced training samples across all sulci, with 91% average overlap within 4mm of the human target. Generalized group results, as expected, showed less overlap with the original human targets, but still performed on average with 80% overlap. Quantitatively, the group results were nonetheless for the most part quite acceptable to an independent human evaluator. The parametrized binary weight description that drives the Dijkstra path optimization is presented as a mechanism to succinctly and compactly describe individual human sulci and groups of sulci.
BibTeX
@misc{pienaar2020sulcus,
  author = {Rudolph Pienaar and Christian Hasselgrove and Kiho Im and David N. Kennedy and P. Ellen Grant and Denise Boriel and Lena Tang and Nikos Makris},
  title = {{“Sulcus Sink”: A Compact Binary and Semi-Automated Inverse Dijkstra-based System for Describing Sulcal Trajectories}},
  year = {2020},
  doi = {10.1101/2020.02.18.955096},
  howpublished = {bioRxiv},
}
Figure