Beyond Role-Based Surgical Domain Modeling

Abstract

Surgical data science increasingly relies on role-based domain models. However, these fail to capture the significance of individual team members and their collaborative dynamics on surgical outcomes. We propose a paradigm shift towards staff-centric surgical domain models, which model individual traits as opposed to considering staff as interchangeable surgical roles. To achieve this, we address the necessary problem of person re-identification in the operating room (OR), which has been hindered due to the challenging visual environment, where traditional biometric cues are obscured. To overcome monotonous texture appearances due to standardized attire, we introduce a novel approach that leverages 3D shape and articulated motion cues to achieve robust, invariant biometric signatures for personnel re-identification.

Beyond Role-Based Surgical Domain Modeling: A Staff-Centric Approach

TL;DR:

We challenge the traditional, role-based view of the operating room and propose a "staff-centric" model that recognizes the unique impact of each individual on surgical dynamics.

Surgical data science has aimed to optimize operating room (OR) workflows by analyzing the roles of the surgical staff. However, this approach treats team members as interchangeable parts, failing to capture the nuanced differences between them.

Given that factors like team familiarity and individual habits significantly affect surgical outcomes, from operative times to complication rates, we argue that this role-based view is not enough. We propose a fundamental shift towards a staff-centric model where each person is recognized as a unique individual, not simply an abstract role.

To see the limitation of the role-centric model, we show two different teams performing the same surgery with the same roles. Notice how their coordination and use of the space differ, highlighting how much team dynamics can change even when roles remain the same.

Team 1

Team 2

Comparison: Team 1 vs Team 2 performing the same surgery with identical roles.

Challenges in Surgical Operating Rooms

The surgical OR is a challenging environment compared to general environments where individuals are easily distinguished by their clothing and faces. Team members wear standardized smocks and gowns to maintain sterility. This homogeneity makes traditional appearance-based identification methods fail.

Standardized attire and protective gear erase the features traditional models rely on to differentiate people.

Generalizable Re-Identification with Geometry

Deep neural networks have a strong bias towards texture. In the OR, these networks learn spurious correlations like street shoes or distinct eyewear visible beneath gowns. When these artifacts vanish in realistic settings, the models fail.

Saliency maps showing how models fixate on simulation artifacts rather than robust biometric features.

We shift our focus to geometry and articulated motion. Our approach encodes personnel as sequences of 3D point clouds to disentangle identity-relevant shape from appearance confounders.

While surgeons are visually ambiguous in RGB, their distinct heights and shapes are measurable in point clouds.

Methodology & Results

We first segment individuals from fused 3D point clouds, then render multi-view 2D depth maps to feed into a ResNet-9 encoder. Features are aggregated over time to generate identity embeddings.

Our point cloud method outperforms RGB counterparts by a 12% margin in accuracy on authentic clinical data. More importantly, it demonstrates superior generalization across different clinical environments.

t-SNE visualization: Point cloud models structure the latent space by stature rather than attire.

Robust Tracking (TrackOR)

Maintaining persistent identities is difficult when staff frequently leave and re-enter. TrackOR integrates our geometric signature into an online tracking pipeline to handle temporary absences and occlusions.

3D Activity Imprints

We propose 3D activity imprints to visualize how individuals utilize the OR space. These imprints provide insight into the coordination of surgical teams and usage patterns for a given surgery.

Head Surgeon 1

Head Surgeon 2

Temporal Pathways

Building on persistent trajectories, we develop Temporal Pathways to analyze dynamic visualizations revealing the specific path staff members take throughout the OR over time.

Temporal pathway imprints revealing potential safety concerns in sterile field proximity.

From Role-Based Surgical Domain Modeling
To Personalized OR Intelligence