Implementation of facial recognition with Microsoft Kinect v2 sensor for patient verification

Area of Science:

• Biomedical Engineering
• Computer Science
• Radiotherapy

Background:

• Accurate patient identification is crucial in radiotherapy to prevent errors.
• Traditional identification methods can be prone to errors.
• Facial recognition offers a potential automated solution for patient verification.

Purpose of the Study:

• To implement and evaluate a facial recognition system for patient identification in radiotherapy.
• To utilize the Microsoft Kinect v2 sensor for capturing facial data.
• To assess the accuracy and feasibility of the system in a clinical context.

Main Methods:

• Developed a facial recognition algorithm using the Kinect v2 Software Development Kit (SDK) and its facial mapping library.
• Extracted 31 fiducial facial landmarks to create a reference dataset.
• Tested the system with a database of 39 faces, performing one-to-one matching and analyzing ROC curves.
• Evaluated system performance under varying ambient light conditions.

Main Results:

• Achieved a sensitivity of 96.5% and specificity of 96.7% across 5299 trials with optimized parameters.
• Demonstrated a robust methodology for real-time facial verification.
• Identified that data collection and matching averaged approximately 30 seconds per patient.
• Found system performance to be most robust under consistent ambient lighting.

Conclusions:

• Facial recognition using the Microsoft Kinect v2 sensor and SDK is a feasible method for patient identification.
• The current system is accurate but time-consuming, requiring further development for clinical practicality.
• Future iterations could offer a cost-effective and robust supplement to existing patient identification methods.