Skills Classification in Cardiac Ultrasound with Temporal Convolution and Domain Knowledge Using a Low-Cost Probe...

Area of Science:

• Medical technology
• Artificial intelligence in medicine
• Ultrasound imaging

Background:

• Point-of-care ultrasound (POCUS) integration necessitates robust operator proficiency assessment.
• Ultrasound proficiency encompasses image acquisition, interpretation, and integration.
• Objective quantification of psychomotor skills in cardiac POCUS acquisition is lacking.

Purpose of the Study:

• To develop and validate an inexpensive system (Probe Watch) for quantifying probe motion during cardiac POCUS.
• To create a machine learning model for classifying sonographer skill levels based on probe motion data.
• To objectively assess the learning curve and skill acquisition in cardiac POCUS.

Main Methods:

• An inexpensive system, Probe Watch, utilizing an inertial measurement device and open-source software, was developed to record probe motion.
• A temporal convolutional network integrating clinical domain knowledge and data augmentation was designed for skill classification.
• The system and assessment method were validated on novice and expert sonographers in a simulation-based cardiac ultrasound training environment.

Main Results:

• The Probe Watch system successfully quantified probe motion during cardiac ultrasound acquisition.
• The temporal convolutional network achieved high accuracy in classifying participants as novice or expert (AUC of 0.931 for snippets, 0.761 for trials).
• Integration of domain knowledge improved the neural network's performance, and key discriminative features were identified.

Conclusions:

• Probe Watch offers an objective method to quantify cardiac POCUS probe motion and assess operator skills.
• This technology can provide valuable insights into sonographer learning curves and skill development.
• Probe Watch has the potential for automated, objective monitoring during POCUS training.