Paper to Pixels: Enhancing Unilateral Neglect Assessment Using the New Computer Vision-Based Tool CANDO

Area of Science:

• Neuroscience
• Computer Science
• Medical Diagnostics

Background:

• The Behavioural Inattention Test (BIT) and Neglect Test (NET) are standard tools for diagnosing neglect.
• Manual scoring of these tests can be inconsistent due to subjective judgment and system limitations.

Purpose of the Study:

• Introduce a novel computer vision tool for automatic scoring of BIT and NET subtests.
• Evaluate the diagnostic validity and efficacy of this automated scoring tool.

Main Methods:

• Manual scoring consistency was assessed across raters with varying expertise.
• A computer vision tool (CANDO) was developed for automated scoring using a deterministic pipeline.
• CANDO's performance was compared against ground truth for cancellation, line bisection, and copying tasks.

Main Results:

• Manual scoring showed inconsistencies in diagnostic classification across raters and assessments, particularly in drawing and copying tasks.
• The CANDO tool achieved high accuracy in cancellation and line bisection tasks and strong agreement in copying tasks.
• Automated scoring reduced variability associated with subjective judgment, time pressure, and oversight, with minimal impact on diagnostic outcomes.

Conclusions:

• Manual assessment of unilateral neglect is prone to inconsistencies.
• Computer vision-based automation enhances the reproducibility and efficiency of neglect assessment.
• The CANDO tool offers a practical solution for higher-quality neglect diagnosis.