Related Experiment Video
Updated: Sep 28, 2025
07:03
Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration
Published on: February 23, 2017
7.8K
Towards Scale and Position Invariant Task Classification using Normalised Visual Scanpaths in Clinical Fetal
Clare Teng1, Harshita Sharma1, Lior Drukker2
1Institute of Biomedical Engineering, University of Oxford, Oxford, UK.
Simplifying Medical Ultrasound : Second International Workshop, ASMUS 2021 : Held in Conjunction with MICCAI 2021, Strasbourg, France, September 27, 2021 : Proceedings. ASMUS (Workshop) (2Nd : 2021 : Online)
|April 4, 2022
Summary
This study introduces a novel method using eye-tracking data to classify tasks during fetal ultrasound scans. The new approach normalizes visual scanpaths, achieving an 84% F1-score and improving upon existing methods for ultrasound task identification.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Human-Computer Interaction
Background:
- Fetal ultrasound imaging consistency varies due to patient factors and operator skill.
- Classifying sonographer tasks is crucial for quality control and training.
- Existing methods struggle with variations in imaging planes and scanning experience.
Purpose of the Study:
- To develop a scale and position invariant method for classifying tasks in fetal ultrasound scans.
- To leverage sonographer eye-tracking data (scanpaths) for automated task recognition.
- To improve the accuracy and robustness of ultrasound task classification.
Main Methods:
- Utilized eye-tracking data to capture sonographer visual attention as scanpaths.
- Developed a normalization technique using bounding boxes to make scanpaths invariant to position and scale.
- Trained machine learning models on normalized scanpath sequences to discriminate between ultrasound tasks.
- Compared the proposed method against approaches using raw eye-tracking data.
Main Results:
- The proposed method using normalized visual scanpaths achieved an 84% F1-score.
- The best performing model significantly outperformed existing methods.
- Demonstrated the effectiveness of scanpath normalization for robust task classification.
Conclusions:
- The developed method offers a reliable way to classify tasks in fetal ultrasound using eye-tracking.
- Scanpath normalization enhances the accuracy and consistency of automated ultrasound task identification.
- This approach has the potential to improve sonographer training and diagnostic efficiency.
