Abstract Archives of the RSNA, 2013
Thomas Sanzalone, Presenter: Nothing to Disclose
Olivier Rouviere MD, Abstract Co-Author: Nothing to Disclose
Caroline Tilikete, Abstract Co-Author: Nothing to Disclose
Yves Berthezene MD, PhD, Abstract Co-Author: Nothing to Disclose
To assess an eye-tracking device recording eye movements during interactive scrolling of stacked CT images.
The study used 10 CT examinations of chest, abdomen and pelvis acquired with a similar protocol in patients with melanoma. Seven CTs were abnormal, one with more than 20 obvious metastases and six with 1 to 3 metastases considered difficult to detect by experts. Nine metastases (“target lesions”) were prospectively selected from these 6 CTs.
74 radiologists interpreted 1 to 3 examinations randomly chosen from the 10 CTs, using a dedicated DICOM-viewer. They were free to scroll up and down through the images. Eye movements were recorded using a remote eye-tracker and synchronized with the image scrolling. 190 interpretations were registered.
Regions of interest (ROIs) were then placed on the 9 “target lesions” and on 9 equivalent areas on the 3 normal CTs. Dedicated software calculated the percentage of radiologists whose gaze was recorded into the ROIs and the dwell time in them.
The average time for each interpretation was 6 minutes and 48 seconds [2 min 19 sec – 16 min 31 sec]. 45 interpretations with radiologist’s gaze properly recorded during less than 75% of the interpretation length were excluded, leaving 145 valid interpretations with an average percentage of properly recorded data of 90.16%.
On average, 66.2% [13.3% - 100%] of “target lesions” were detected, with an average dwell time of 3.84sec [0.14sec – 7.89sec]. The average percentage of visualization of equivalent ROIs on normal CTs was 40.3% [2% - 90.2%], with an average dwell time of 0.33sec [0.02sec – 0.85sec].
Two metastases were detected by 80% and 100% of radiologists when equivalent normal areas were visualized by only 9.8% and 3% of readers suggesting a role of the peripheral vision in their detection. On the other hand, one poorly-contrasted metastasis, detected by only 50% of radiologists, was located in an area visualized by 90.2% of radiologists on normal CTs.
Eye-tracking during interactive scrolling of stacked images is feasible and compatible with routine CT interpretation. Peripheral vision may play a role in metastases detection, as well as the contrast and location of the metastases themselves.
Eye-tracking studies during CT interpretation could become a useful tool for analyzing factors influencing metastases detection and thereby improving routine practice.
Sanzalone, T,
Rouviere, O,
Tilikete, C,
Berthezene, Y,
Is It Possible to Track Radiologists’ Gaze When They Scroll Through Stacked CT Images? Eye Tracking Analysis of Radiologists' Reading Computed Tomography for Metastases Detection. Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL.
http://archive.rsna.org/2013/13044150.html