In recent years MRI has become more important as an imaging modality for ophthalmology^1,2, especially because it has a higher penetration depth compared to conventional optical approaches. Eye-motion, however, limits its clinical potential, as the resulting artifacts hinder clinical evaluation. Dedicated scanning protocols, such as cued-blinking protocols^1,3,4, are able to reduce but not eliminate these motion artifacts. Experience in clinical practice shows that patients are not always able to fully comply with cued-blinking instructions resulting in incidental blinks during data-acquisition. In this context, we describe the use of field probes (FP) to detect eye blinks. When one or more accidental blinks are detected, imaging data can be corrected (Figure 1) by two methods, a prospective correction by closed-loop k-space reacquisition, and retrospective data correction using Projection Onto Convex Sets (POCS).

Experimental setup: The study was approved by the local institutional review board and six healthy volunteers were included after giving informed consent. All experiments were performed on a 7T whole body system (Philips Healthcare, The Netherlands), equipped with a dual channel transmit head coil (Nova Medical, USA) and an in-house build receive eye coil. The eye blink signal was measured with one field probe (Skope Magnetic Resonance Technologies, Switzerland) placed close to the contralateral eye: the frequency shift produced by blinking was enhanced by the application of a small amount of mascara which contains ferromagnetic iron oxide pigment.

Blink detection and correction was implemented using a FP system which was synchronized to the MR-acquisition. The FP measured the local magnetic field which we hypothesize would change as the eye-lid moves during an eye-blink. The FP data was compared to a pre-determined frequency shift threshold to detect the blinks. If a blink was detected during MR-acquisition, the scanner was instructed to rescan the last block of k-space lines (prospective correction). Retrospective correction was performed by removing the detected corrupted k-lines and performing POCS on the remaining data.

Study protocol: A cued blinking protocol was implemented by means of visual instructions to the volunteer. Two scans were made per volunteer, in the first, the volunteer was asked to deliberately blink twice during acquisition resulting in imaging artifacts. In the second scan, the volunteer was asked to adhere to the cued blinking instructions.

MR acquisition parameters: A Turbo Spin Echo scan (TSE factor 8), normally used in imaging ocular tumors, was used to demonstrate the principle of blink detection and correction. The imaging field-of-view was 44.5 mm x 89 mm, covering the eye and optical nerve in 4 sagittal slices of 1 mm thickness, with an in-plane resolution of 0.35 mm x 0.35 mm2. The scan duration was 45 sec.

An example FP measurement of an eye blink is shown in Figure 2. The mascara increased the FP blink frequency shift by factors of between 5 and 15. The detected blinks showed an average frequency shift of 127 Hz, range [39 Hz to 228 Hz]. For the six volunteers studied the FP setup successfully detected all the blinks which occurred during the MR signal readout, which subsequently triggered the scanner to reacquire the corrupted data. Figure 3 shows the uncorrected and resulting retrospective and prospective corrected images for 3 subjects together with a scan without blinks for reference.

Eye blinks during ocular imaging can successfully be detected by a simple field probe. The probe can be integrated into the eye coil, which has the advantage in terms of patient comfort that no contact between the probe and the patient is necessary.

The quality of high-resolution eye images acquired at 7T improves after correction for FP-detected eye blinks, both with prospective and retrospective approaches. The improvement is greater for the triggered reacquisition method, which has the added advantage that the images can be directly reconstructed on the scanner using the standard reconstruction algorithms of the scanner. Comparisons of the sensitivity of this approach with alternative methods, such as detection using ECG electrodes, remains to be performed, but this very simple method can be easily extended to other MR readouts and sequences such as EPI or localized spectroscopy.