In this ongoing project, funded by NIH, we are aiming to collect eye-tracking data as a non-intrusive way of providing localization information on labels associated with each chest x-ray read by radiologists. We are collecting data for a subset of images from the MIMIC-CXR dataset. The data collection includes the audio recording of reports, enabling a timestamped association between the location of eye fixations and expressions dictated. We are also collecting global labels, bounding boxes for anomalies, certainty, and a bounding box around lungs and heart. The project includes:

• analyzing the alignment between dictation / eye fixations / pupil dilation / bounding box annotations,
• providing all collected data as an open dataset, and
• using collected data as supervision for novel deep learning models.

The image below is an example of a data-collection screen, where radiologists can select an ellipse around anomalies, providing a certainty for that label, while eye-tracking data is collected. A frontal chest x-ray is shown in the screen's center, while the screen's margins show buttons and additional information. The GitHub repository containing the code for the MATLAB interface can be found at https://github.com/ricbl/eyetracking



A link to download the dataset is expected to be available in June 2021.

Contacts

PI: Tolga Tasdizen, This email address is being protected from spambots. You need JavaScript enabled to view it.
Research assistant: Ricardo Bigolin Lanfredi, This email address is being protected from spambots. You need JavaScript enabled to view it.
Lead radiologist: Joyce D. Schroeder, This email address is being protected from spambots. You need JavaScript enabled to view it.