Fast Adaptive Optics for 3D Medical Imaging

News

Not a member yet? Register for full benefits!

Real-time medical imaging is a long-time goal of medical VR systems. The ability to project a virtual image of the patient, or the area of the patient of interest through a system at the same time as the patient is being scanned – say, by a scanner in the operating theatre itself, would be a great boon to many procedures. If the test the surgeon needs wasn't done, do it now. Immediate results.

It could even open the way for live, streaming data detailing the patient's condition, continually updating as the surgeon operates. We have such virtual data already, but they are virtualised results of a prior scan of the patient, overlaid on the vision of the surgeon as they operate now.

However, if we are to achieve real-time medical scan data virtualisation, there are still a great many hurdles which must be first overcome. One of these, a big one, is the distortion that occurs on medical scan data when the patient moves (breathes or has their heart beat) during the scanning process. As stopping the heart or diaphragm from operating during a scan has some undesirable side effects – such as killing the patient - other ways of dealing with distortion must be found.

Distortion is a problem in CT and CAT scans notably. This is because of how they operate. Both are tomographic scans. They section the patient in fragments of a slice at a time, then when one slice is completed they move on to the next. This is where the problem occurs. When a patient breathes or their heart inflates, the body parts around these organs move subtly. This means that landmarks are in a slightly different position from one part of a slice to another. So, when the slice is assembled, it doesn't quite fit together as it should. Then when the next slice is done, the patient has been moving throughout, and it doesn't quite line up perfectly with the previous one – and so on.

The differences are minor – streaks rather than points – but when you are going to be using the data provided, to cut the insides about with it as a reference, having as high a fidelity as possible is most definitely desirable.

The image above shows standard OCT data (Optical Coherence Tomography), however similar results are achieved with all tomography-based systems. As explained above, its just a side-effect of the imaging method when applied to living systems.

The image is of gel-based phantoms laced with microparticles, but you would be hard-pressed to tell that, from all the streaking and blurring present. What's needed to make this data useful, is some form of corrective routine. Such routines are routinely carried out, to clear up the image, reduce distortion and make features recognisable. However, they are much like any other multi-pass rendering technique. They take quite some time to run, and the finished image is available still by still, with each computed separately. It works wonderfully, but real-time medical imaging will never be possible when we are waiting anything from five minutes to half an hour per frame for the correctional routines to finish re-rendering.

Active research is attempting to tackle the problem. Like anything else in real-time virtual reality, you have to discard many of the techniques in used in passive VR in order to get the results you seek. Often what passive VR does at the software level, you need to do at the hardware level to be fast enough for real-time work.

University of Illinois researchers believe they may have done it, by looking at how astronomy uses adaptive optics to correct their photos – which suffer similar effects due to the relative motion of the Earth, to the systems they are studying. In astronomy a complex system of mirrors smooth out the scattered light before it enters the lens. That is not really possible in tomographic systems, although it has been tried before. Hardware based adaptive optics are complicated, tedious to align and extremely expensive. They can only focus on one focal plane at a time, so for tomography the mirrors have to be adjusted and a new image scanned for each focal plane. It makes the scan much slower for the patient, and the machines much bulkier. In addition, how do you add corrective mirrors to a hand-held optic probe? What about for a retinal scanner?

Mirrors are utterly impractical for small, fast medical imaging. However, what we can do is look at how the mirror system works, then apply that processing to the image after it has been taken, as a single dedicated rendering pass. Quick, computationally cheap, and it works.

One of the researchers, Stephen Boppart, a professor of electrical and computer engineering, of bioengineering and of internal medicine at the university, had this to say: “It’s the same challenge, but instead of imaging through the atmosphere, we’re imaging through tissue, and instead of imaging a star, we’re imaging a cell. But a lot of the optical problems are the same.”

Here we can see two corrected images, one using the single-pass render filter on the distorted OCT image above, the second an ISAM image (Interferometric Synthetic Aperture Microscopy) which is itself a variation on OCT. It is included to show the correction method works, regardless of the topographic method used.

Another of the researchers, Steven Adie, a postdoctoral researcher at the Beckman Institute for Advanced Science and Technology stated that “Computational techniques allow you to go beyond what the optical system can do alone, to ultimately get the best quality images and three-dimensional datasets. This would be very useful for real-time imaging applications such as image-guided surgery.”

For example, computational adaptive optics could be very useful for ophthalmologists. Boppart’s group previously has developed various handheld optical tomography devices for imaging inside the eye, particularly retinal scanning. Aberrations are very common in the human eye, making it difficult to acquire clear images. But adaptive optics hardware is too expensive or too complicated for most practising ophthalmologists. With a computational solution, many more ophthalmologists could more effectively examine and treat their patients.
“The effectiveness is striking,” Boppart said. “Because of the aberrations of the human eye, when you look at the retina without adaptive optics you just see variations of light and dark areas that represent the rods and cones. But when you use adaptive optics, you see the rods and cones as distinct objects.”

The work has been described in the Proceedings of the National Academy of Sciences.