Time to dust off the inertia and expound on subjects of the Brain, Spine and Neurosurgery. WordPress sent a friendly reminder that it had updated itself and my domain provider also sent a similar friendly request for the annual subscription. It then led to me starting to write this post.
We have options of Pediatric Neurosurgery (often heartbreaking and sometimes incredibly rewarding), Minimal Access Surgery and a conference and talks given about the same, and the latest kid on the block Robotics and AI.
The buzzwords of today are artificial and intelligent. Throw in Robotics for good measure. Surgical Robots today have shown substantial improvement with each passing generation to make surgery safer, more precise and to reduce human error and fatigue. Neurosurgery is no stranger to technology and it would seem that robotics have taken to neurosurgery fairly well. But what do these robots do. They aren’t like the medical droids in Star Wars, and we’re nowhere near the complexity of a tricorder from Star Trek. The current set of robots are effectively robotic arms. They are integrated in to the operative microscope, the neuronavigation system, or are standalone robots with their own cameras and navigation.
Currently (and maybe thankfully) the available robots need to be told what to do. What they do right now is position themselves along predetermined surgical corridors to ensure accuracy in reaching a particular target.
Let’s break that down with an example of a stereotactic brain biopsy. This is where a small area of abnormality which is deep in the brain can be accessed via a needle through a small hole in the skull. The reason to do this is commonly a tumor which is in a location that may be too deep or too risky to approach in the traditional neurosurgical technique. Tumors closer to the surface are usually approached by opening a window in the skull and removing the tumor with the aid of a surgical operating microscope. The trouble of course, is how do we know we are at the target from outside the rather opaque skull. The answer was applying Descartes’ principle. Not the one of thinking, therefore being, but the one of Cartesian planes and coordinates. This basically means that if we assume a point where 3 perpendicular lines meet and assume this point to be 0,0,0 in the x,y, and z axes, then any point in space can be described by a set of coordinates relative to the 0,0,0 point. How we do that with the brain is by fixing the patient’s head into what looks like (depending on your age and love of pop culture) either a Nazgul, The man in the iron mask or Bane from the batman movies. This device is a sterotactic frame. My favourite one is the Leksell frame, but there are others. The actual math and technique is rather exhaustive to write out but to summarize, we fix the frame onto the patient’s head under local or general anesthesia and get a scan with the frame in place. Special markers and then some nifty software mumbo-jumbo allows us to then define 2 points. One target point. And a surface entry. What we then get is a line or a trajectory and a depth to reach said target. Simple? The images below should help clear this up.
As is fairly evident, this is a cumbersome process fraught with potential issues. The next development that occurred was frameless or navigated stereotactic biopsy. Here the scan of the patient is fed into the navigation machine and the target and entry are similarly determined. A stereotactic wand is then introduced and the navigation machine then tells us, in real time if we are in the right entry point and have the right trajectory. The next step is to get a biopsy needle that is also recognized by the navigation that introduce it through a small hole in the skull (the entry point), and pass it along the predetermined trajectory. The navigation of course gives us real time feedback on if we are on target or have deviated and if so by how many millimeters. The biopsy needle is then advanced till the exact depth to reach the target, again with real-time imaging feedback and a biopsy taken.
The next advancement is if there is a robotic arm that can be seen and guided by the navigation then wouldn’t it be more accurate than a human arm and human eyes. Which is the constant endeavor of advancing medical technology. These robotic arms now exist and have made the chances of error less than 1mm. The procedure however remains the same. We still need to identify patients who need a biopsy. We still need to determine the targets and safe entry points and safer trajectories. But the robot ensures the accuracy of the procedure. The common one used for this type of work is an innovatively named cute little fellow called the AutoGuide. It looks like so.
It fixes on to the operating table and can align itself based on the plan created on the navigation system. A series of steps with dedicated drills, biopsy needles, all of which are passed through a rigid port in the robot makes the entire workflow a breeze. In situations where multiple electrodes need to be placed in the brain, this is particularly useful since it can maneuver itself to each position easily instead of dealing with the cumbersome Leksell frame. The reason to do this science fictionesque procedure is to locate which set of brain cells maybe short circuiting and causing repeated seizures.
As is apparent, robots in neurosurgery are effectively arms that can be guided by navigation systems to position themselves in space to ensure that the surgery is performed in the exact trajectory that is determined by the surgeon. Nowhere is this more challenging than in spine surgery. A lot of spine surgery nowadays involves the use of implants that go into the vertebral bones in order to confer stability to the spine. These screws have traditionally been inserted based on anatomical landmarks that are visible during surgery. And the passage of the screw is at least partly based on tactile feedback and the surgeon’s experience. The images below show how screws are normally inserted in the lumbar (lower back) or cervical (neck) spine. (Courtesy AOSpine)
This seems fairly straight forward till you realise that the only part of the vertebra that is visible to the surgeon is the posterior third. And the screw is effectively inserted blind. in the lower image above, the red circle outlines the zone of doom, so to speak. the bone is about 4 to 5mm thick, the screw is 3.5 to 4mm in diameter. One side has the spinal cord, the other is the vertebral artery (only 1 of the 4 arteries that supply the brain), above and below the screw lie the nerves that run to the arm. Needless to day, Neurosurgery and spine surgery is an extremely delicate and precise undertaking. One way of reducing the chances of a screw malposition is to use an intraoperative CT and a navigation. Like brain navigation, spine navigation pretty much allows surgeons to plan a safe trajectory for the screw based on the CT of the patient and insert the screws precisely with little risk of damage. Now by extension, we can use a robot to make that process even safer, and more accurate.
The Mazor X, Rosa, Renaissance systems are currently available Spine robotic platforms that do just this.
Now, where does AI or at least Machine Learning, come in? Some platforms are using machine learning to outline tumors in MRI scans, based on how different they are from normal tissue. This does sometimes need correction by the surgeon but the system more or less gets it right. Given data about which parts of the brain are important or eloquent (as in they do something that we can tangibly qualify) and what structures (blood vessels, ventricles) to avoid, a safe trajectory can also be plotted. The plan just needs to be confirmed and it would take the guesswork out of planning surgeries. Similarly with spine, the existing robots can populate screws on to the CT images of the patient given the surgical plan of maybe how many levels require fusion. They can pick the correct sizes, trajectories and alignments. Newer systems also allow for more complex measurements. In scoliosis where the spine is curved, correction often involves putting screws into very deformed vertebrae, which in itself is a challenge. And the next step is manually bending a rod in a way to fit along to the curved spine. To date, this is done by effectively eyeballing the spine and requires a huge amount of experience to get things right. This is also what makes Scoliosis and deformity surgery one of the most difficult branches of spine surgery. The current robotic systems allow for preoperative planning of all the screws and the planned correction and then, the rod curvature is determined by the machine. That curvature can be factory created to the precise specifications again eliminating human error.
Robots and AI are here to stay and possibly the only barrier to their widespread utilization is their cost. They reduce human error and operator fatigue and have multiple internal and external checks and balances to ensure we don’t have a Skynet type situation. Every passing week upgrades the capabilities of generative AI as is evident in this post, a lot of which has been co authored by ChatGPT.
Will AI replace the doctor? Not anytime soon, it may make our lives more difficult if Dr Google and Dr GPT join hands to convince patients that they have Lupus, when they have a rash secondary to an allergy to rouge. We’ll have to wait and see.