The effect of expertise on eye movement behaviour in medical image perception

Abstract

The present eye-movement study assessed the effect of expertise on eye-movement behaviour during image perception in the medical domain. To this end, radiologists, computed-tomography radiographers and psychology students were exposed to nine volumes of multi-slice, stack-view, axial computed-tomography images from the upper to the lower part of the abdomen with or without abnormality. The images were presented in succession at low, medium or high speed, while the participants had to detect enlarged lymph nodes or other visually more salient abnormalities. The radiologists outperformed both other groups in the detection of enlarged lymph nodes and their eye-movement behaviour also differed from the other groups. Their general strategy was to use saccades of shorter amplitude than the two other participant groups. In the presence of enlarged lymph nodes, they increased the number of fixations on the relevant areas and reverted to even shorter saccades. In volumes containing enlarged lymph nodes, radiologists' fixation durations were longer in comparison to their fixation durations in volumes without enlarged lymph nodes. More salient abnormalities were detected equally well by radiologists and radiographers, with both groups outperforming psychology students. However, to accomplish this, radiologists actually needed fewer fixations on the relevant areas than the radiographers. On the basis of these results, we argue that expert behaviour is manifested in distinct eye-movement patterns of proactivity, reactivity and suppression, depending on the nature of the task and the presence of abnormalities at any given moment.

Figure 1. Cross-sectional abdominal CT-images used in the experiment.

The CT-images depict an enlarged lymph node from an ELN-video (arrowhead in a) and liver cysts from an ABN-video (yellow lines in b). Note that the left and right side of the body are switched in the images.

Figure 2. Cross-sectional abdominal CT-image with an ELN divided into 20 equal-sized areas of interest (AOIs).

The enlarged lymph node in this image is pointed out by the black arrowhead and resides in Area 10, the area where most lymph nodes are located in all images across videos.

Figure 3. Heatmaps of the distribution of fixations over all 20 areas for each participant group.

The left panels include all fixations over all videos; the right panels depict the fixation distributions for images in ELN videos where enlarged lymph nodes are present.

Figure 4. Cross-sectional abdominal CT-image depicting hydronephrosis.

The hydronephrosis is present at the right side in area 2, 3, 6, and 7 and on the left side in area 15.

Figure 5. Interaction between Expertise and Video for average fixation duration.

The figure depicts that experts reverted to longer fixation durations in ELN-videos than in other videos, whereas semi-experts and naïve participants did not.

Figure 6. Interaction between Expertise and Video for saccadic amplitude.

The figure depicts that semi-experts and to some extent naïve participants used longer saccades in ABN-videos than in other videos, whereas experts did not.

Figure 7. Interaction between Expertise and Presence of ELN for saccadic amplitude in ELN-videos.

The figure depicts that unlike the other groups, experts made shorter saccades in video sections where enlarged lymph nodes were present than in sections where they were absent.

Figure 8. Interaction between Expertise and Framerate for saccadic amplitude in ELN videos.

Experts made progressively longer and naïve participants progressively shorter saccades as a function of increasing framerate. Semi-experts behaved like naïve participants in the lowest framerate and like experts in the higher framerates.

Figure 9. Interaction between Expertise and ABN for saccadic amplitude in ABN videos.

Unlike experts, both semi-experts and naïve participants reverted to clearly longer saccadic amplitudes in video sections where abnormalities were present.