| Summary: | <p>Fetal sonography remains a highly specialised skill in spite of its necessity and importance. Because of differences in fetal and maternal anatomy, and human pyschomotor skills, there is an intra- and inter-sonographer variability amoungst expert sonographers. By understanding their similarities and differences, we want to build more interpretive models to assist a sonographer who is less experienced in scanning.</p>
<br>
<p>This thesis’s contributions to the field of fetal sonography can be grouped into two themes. First I have used data visualisation and machine learning methods to show that a sonographer’s search strategy is anatomical (plane) dependent. Second, I show that a sonographer’s style and human skill of scanning is not easily disentangled.</p>
<br>
We first examine task-specific spatio-temporal gaze behaviour through the use of <em>data visualisation</em>, where a <em>task</em> is defined as a specific anatomical plane the sonographer is searching for. The qualitative analysis is performed at both a <em>population</em> and <em>individual</em> level, where we show that the task being performed
determines the sonographer’s gaze behaviour.</p>
<br>
<p>In our <em>population</em>-level analysis, we use unsupervised methods to identify meaningful gaze patterns and visualise task-level differences. In our <em>individual</em>-level analysis, we use a deep learning model to provide context to the eye-tracking data with respect to the ultrasound image. We then use an event-based visualisation to understand differences between gaze patterns of sonographers performing the same task.</p>
<br>
<p>In some instances, sonographers adopt a different search strategy which is seen in the misclassified instances of an eye-tracking <em>task classification</em> model. Our <em>task classification model</em> supports the qualitative behaviour seen in our <em>population</em>-level analysis, where task-specific gaze behaviour is quantitatively distinct.</p>
<br>
<p>We also investigate the use of time-based skill definitions and their appropriateness in fetal ultrasound sonography; a time-based skill definition uses years of clinical experience as an indicator of skill. The developed task-agnostic <em>skill classification</em> model differentiates gaze behaviour between sonographers in training and fully qualified sonographers. The preliminary results also show that fetal sonography scanning remains an operator-dependent skill, where the notion of human skill and individual scanning stylistic differences cannot be easily disentangled.</p>
<br>
<p>Our work demonstrates how and where sonographers look at whilst scanning, which can be used as a stepping stone for building style-agnostic skill models.</p>
|
|---|