Magnetic resonance-based eye tracking using deep neural networks
DeepMReye tool is open source app: https://github.com/DeepMReye/DeepMReye/
The researchers also worked hard to make the tool as user friendly as possible, and created a website with user recommendations and a FAQ: https://deepmreye.slite.com/p/channel/MUgmvViEbaATSrqt3susLZ
To record eye movements, research institutions typically use a so-called eye tracker - a sensor technology in which infrared light is projected onto the retina, reflected, and eventually measured. "Because an MRI has a very strong magnetic field, you need special MRI-compatible equipment, which is often not feasible for clinics and small laboratories", says study author Matthias Nau, who developed the new alternative together with Markus Frey and Christian Doeller. The high cost of these cameras and the experimental effort involved in their use have so far prevented the widespread use of eye tracking in MRI examinations. That could now change. The scientists from Leipzig and Trondheim developed the easy-to-use software "DeepMReye" and provide it for free.
With it, it is now possible to track participants’ viewing behavior even without a camera during an MRI scan. "The neural network we use detects specific patterns in the MRI signal from the eyes. This allows us to predict where the person is looking. Artificial intelligence helps a lot here, because we often don't know exactly which patterns to look for as scientists", Markus Frey explains. He and his colleagues have trained the neural network with their own and publicly available data from study participants in such a way that it can now perform eye tracking in data the software has not been trained on. This opens up many possibilities. For example, it is now possible to study the gaze behaviour of participants and patients in existing MRI data, which were originally acquired without eye tracking. In this way, scientists could use older studies and data sets to answer entirely new questions.
The software can also predict when eyes are open or closed. Moreover, it can track eye movements even when the eyes remain closed. This may allow to perform eye tracking even when study participants are asleep. "I can imagine that the software will also be used in the clinical field, for example, in the sleep lab to study eye movements in different sleep stages", says Matthias Nau. In addition, for blind patients, the traditional eye-tracking cameras have rarely been used because an accurate calibration was very cumbersome. "Here too, studies can be carried out more easily with DeepMReye, as the artificial intelligence can be calibrated with the help of healthy subjects and then be applied in examinations of blind patients." The software could thus enable a variety of applications in research and clinical settings, perhaps even leading to eye tracking finally becoming a standard in MRI studies and everyday clinical practice.
Viewing behavior provides a window into many central aspects of human cognition and health, and it is an important variable of interest or confound in many functional magnetic resonance imaging (fMRI) studies. To make eye tracking freely and widely available for MRI research, we developed DeepMReye, a convolutional neural network (CNN) that decodes gaze position from the magnetic resonance signal of the eyeballs. It performs cameraless eye tracking at subimaging temporal resolution in held-out participants with little training data and across a broad range of scanning protocols. Critically, it works even in existing datasets and when the eyes are closed. Decoded eye movements explain network-wide brain activity also in regions not associated with oculomotor function. This work emphasizes the importance of eye tracking for the interpretation of fMRI results and provides an open source software solution that is widely applicable in research and clinical settings.