Visualising Spatio-Temporal Gaze Characteristics for Exploratory Data Analysis in Clinical Fetal Ultrasound Scans.

Area of Science:

• Medical Imaging
• Ophthalmology
• Data Visualization

Background:

• Visualizing eye movements during fetal ultrasound interpretation is difficult due to varying Areas-of-Interest (AOIs).
• Current AOI identification methods are manual or manufacturer-specific, lacking study-specific relevance.

Purpose of the Study:

• To develop an unsupervised clustering method for identifying meaningful AOIs in fetal ultrasound videos.
• To visualize spatio-temporal gaze characteristics using bi-contour plots.
• To compare the proposed method with standard eye-tracking algorithms.

Main Methods:

• Utilized Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) for unsupervised AOI identification.
• Employed corresponding images to capture granular changes within identified AOIs.
• Visualized transitions within and between AOIs based on sonographer gaze patterns.

Main Results:

• The proposed method successfully identified meaningful AOIs without manual labeling.
• HDBSCAN captured granular changes in gaze characteristics not revealed by standard methods.
• Bi-contour plots effectively visualized spatio-temporal gaze patterns.

Conclusions:

• The unsupervised clustering approach offers a novel and effective way to analyze eye-tracking data in medical imaging.
• This method provides deeper insights into clinicians' visual attention during fetal ultrasound interpretation.
• The technique is well-suited for exploratory analysis of complex eye-tracking datasets with multiple participants and AOIs.