By Saliou Thiam, Fall 2023.
Since neurosurgery was invented in the early 1900s by Harvey Cushing, there has been monumental progress made in the realm of improving neurological surgical techniques and instrumentation to better the outcomes of patients. In 2023, there are now countless different procedures and methods surgeons have in their arsenal to deal with tumors that are extremely complex and are in difficult-to-reach portions of the brain. One aspect of neurosurgery that hasn’t seen such progression is the understanding of the molecular profile of brain tumors. Molecular profiling of tumors — identifying specific markers that give us information on the subtype of tumors — plays a critical role in diagnoses. Many molecular components of tumors themselves are actually still a mystery, not only to us as the general public, but even to physicians and neuroscientists themselves.
During neurosurgery, surgeons tasked with tumor resection often have a very difficult choice to make. That choice is to either resect the tumor while cutting some of the surrounding healthy brain matter or resect part of the tumor, leaving residuals that will grow again over time. There is also the possibility that the surgeon is able to cleanly remove the tumor without any margins or residuals, but this case is extremely rare as brain tumors embed themselves into the surrounding healthy gray and white matter of the brain. This is an excruciating decision… as on one hand there is the possibility of permanent brain damage, and on the other hand, the patient still has cancer and will need another surgery down the line. Scientists in the Netherlands are reporting that through using artificial intelligence during surgeries, they are able to gain more information about the tumor that may help them make this decision.
Our current and most widely used methods of analyzing tumor samples involve examining the sample under a microscope and sending it to a genetic sequencing facility where they are able to gain more information on the molecular profile of the tumor itself. This process takes several weeks to receive the results back. Considering that brain tumors are extremely aggressive, many patients do not have weeks to wait until the results are back, causing surgeons to “start treatment without knowing what [they’re] treating” [1]. Without all of the information needed to create an extensive treatment plan, though it does not happen often, it is possible for surgeons to approach a tumor with an overly aggressive approach relative to the type of tumor they are dealing with. This can have serious implications for the patient as margins that are too wide may result in excess brain matter being removed from the patient’s brain.
This new integrative AI technology plans to help surgeons identify the specific subtype of tumor, thus providing crucial information in determining the proper treatment plan. This cutting-edge technology involves a computer scanning certain segments of a tumor’s DNA and highlighting certain chemical modifications that will help in the detailed diagnosis of the tumor’s subtype. This entire process would occur during the surgery within ninety minutes of the sample being resected from the patient. Having this information will allow surgeons to be better equipped to decide how aggressive to be during hour-long surgeries.
Sturgeon, the name of this AI technology, was initially given frozen tumor samples in order to test its accuracy with its diagnoses. Out of fifty frozen tumor samples, Sturgeon was able to diagnose forty-five out of fifty samples accurately, and with the remaining five, refrained from giving a conclusive diagnosis because the information provided was unclear. After this section, live brain samples were used during surgeries. Out of twenty-five surgeries, Sturgeon was able to correctly diagnose eighteen tumors, and in the other seven cases, refrained from giving a conclusive diagnosis. Given the trials the AI has already gone through, it is important to note the fact that Sturgeon has never given a false diagnosis. It is up to par with what is expected of surgical interfaces. If anything, there seems to be an overall benefit to having Sturgeon in the operating room, as it will be able to provide accurate results, and in the case where the results are inconclusive, then surgeons would simply rely on current scientific methods to complete the surgery.
Though it has the potential to be helpful, Sturgeon itself still has some drawbacks preventing it from becoming a mandatory part of every operating room. Firstly, when obtaining a sample for Sturgeon to analyze, it relies on the DNA of the sample, which should be a tumor, to draw its results. In the case that a surgeon is unable to get clean margins and there is some healthy brain matter in the sample, it will prove difficult for Sturgeon to draw accurate conclusions. Secondly, the small sample may have some underlying molecular differences with the rest of the tumor. The small segments that are being sequenced may not be fully representative of the rest of the tumor. Thirdly, this technology can not be applied to other growths outside of brain tumors, as brain tumors are the most suited for classification via chemical modifications. Tumors of the vital organs in the torso are subject to so much variability that it makes it difficult to standardize tumor classification. Fourthly, Sturgeon’s implementation isn’t meant to be an end-all approach to tumor classification during surgery. There is still a need for significant expertise in bioinformatics and individuals who are able to run and repair technology such as this one in high-stress situations.
This new technology is part of a monumental step in the understanding of the molecular profile of tumors. Sturgeon can potentially help scientists develop targeted treatments for specific subtypes of brain tumors that are less damaging to the nervous system, further improving patient outcomes from invasive neurosurgeries.
[1] Johns Hopkins Medicine. “History of Neurology and Neurosurgery.” Accessed [11/9/2023].
[2] The New York Times. “AI Shows Promise in Tumor Diagnosis.” October 11, 2023. Accessed [11/9/2023].