Freehand OCT

Add One

Optical Coherence Tomography (OCT) images are usually built up scanning a laser spot rapidly over the sample. For the common applications of eye and skin imaging  there aren’t any particularly onerous size constraints on OCT systems, so bulky galvanometer mirrors can be used to generate the scan. But for endoscopic imaging we need to minimise the size of the mechanism. This has resulted in the development of a number of miniaturised scanning systems, mostly involving MEMS mirrors and fibre scanning cantilevers. One of the more esoteric solutions is the idea of free-hand scanning, where the operator builds up a scan simply by moving the probe manually. This reduces the hardware requirements to a minimum, but leads to the tricky problem of how to correctly assemble the image when we don’t know how the probe has been moved.

The first paper to demonstrate viable manual scanning was published in 20091. The authors used a cross-correlation method to reconstruct the B-scan. (If you’re unfamiliar with the terminology of scan directions in OCT, have a read of this page). The first recorded A-scan is taken as a reference, and the cross-correlation coefficient is then calculated between this reference and subsequent A-scans. The cross-correlation coefficient essentially tells us how similar the two A-scans are. If the probe has only moved a small distance then the A-scans will overlap and the cross-correlation coefficient will be high. If the distance moved is greater, the overlap will be less, the A-scans will be distinct and the cross-correlation coefficient will be lower. So the coefficient is a measure of the distance that the probe has moved, and can be used to decide on the position of the A-scan in the image.

This method isn’t without its limitations. Firstly, we are relying on the probe only being moved a very small distance between each A-scan. We need this distance to be considerably smaller than the lateral resolution of the system, otherwise there won’t be a smooth fall-off in the correlation coefficient. The relationship between the correlation coefficient and real physical distance is also going to depend on the particular sample we’re looking at. Something that is relatively homogeneous might have high correlation coefficients even when there is no overlap between the beams. So although the images may look right, we have no way of knowing if they are an accurate representation of the structure, and certainly no way of performing an accurate calibration of the lateral image size.

Phase Shifts

The topic didn’t receive much attention again for a little while. But in 2012, a paper appeared in Optics Letters which used a different approach to estimating the lateral motion2. This group made use of the same effect that powers for Doppler  OCT. When the probe to sample distance changes in-between the acquisition of two A-scans then there will be a phase shift in the interference pattern. Providing the sample hasn’t moved then the magnitude of this phase shift is related to the change in the axial position of the probe.

Of course we’re not so interested in the axial motion, it’s the lateral position change we need. The authors’ idea was to hold the probe so that the optical axis isn’t exactly perpendicular to the lateral motion. So when the probe is moved laterally there’s a component of the velocity which is along the optical axis. This will then result in a phase shift which can be measured and used to estimate the velocity of the probe. This information then allows us to correctly assemble the A-scans into an image.

Again, there’s a limit to how fast the probe can be allowed to move. If the phase changes by more than a complete cycle (i.e. one wavelength) between scans then there’ll be a wrapping effect and the velocity will be under-estimated. And unless we know the angle the probe is at, then we can’t perform an absolute distance calibration. So although the method works in principle, there are clear questions over how viable it would be in clinical practice.

Electromagnetic Tracking

At almost the same time, David Sampson’s Group from the University of Western Australia published an alternative approach using electromagnetic trackers. These are small probes which can be tracked in 3D space as long as they are within range of a transmitters/receiver unit. The group had previously used these devices for localisation of OCT endoscopic probes to assist with the reconstruction of 3D volumes3. But conventional wisdom would suggest that magnetic trackers are a totally inadequate tool for allowing manual scanning. In particular, their accuracy is usually quoted as being closer to millimetres than the few microns which would be require to correctly align the A-scans. The key idea of this paper was that the authors showed that it is possible to achieve very high precision with these devices if they’re used an in appropriate way4.

Firstly, unlike in most applications of magnetic tracking, we are not concerned with the global accuracy of the positioning, but only the relative accuracy over displacements of a few millimetres. This is a huge relaxation of the requirements, particularly as one of the main sources of inaccuracy – stray magnetic fields – are likely to be broadly constant over this distance scale.  The group reduced inaccuracies even further by calculating the general trend of the motion at any one time, rather than relying on individual position measurements.

The result was they were able to produce high quality B-scans through a side-firing needle based probe. It’s difficult to know how well this would work for flexible probes, where the positioning is less constrained. But the paper is interesting in that it shows that there is certainly some potential for electromagnetic tracking to interface with optical imaging devices, despite the apparent order of magnitude difference in resolution.


Leave a Comment

Your email address will not be published. Required fields are marked *


You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>