Can surgeons identify aphasia risks when removing a brain tumor? To find out, researchers at Clinicum Rechts der Isar of the Technical University of Munich (TUM) are analyzing the brain as a network. In a current study of 60 patients, they actually achieved an accuracy rate with three-quarters of their expectations.
Brain tumors are relatively rare. According to the German Society of Neurology, the annual incidence is about five cases per 100,000 population. “But in most cases surgical removal of the tumor is unavoidable,” says Professor Sandro Krieg, who estimates that a glioma – a common type of brain tumor – is removed at the Klinikum Rechts der Isar of the Technical University of Munich (TUM). “On an almost daily basis.”
Depending on the tumor, Craig and colleagues develop individualized treatment and surgical strategies. Crucial point: healthy tissue must be preserved as far as possible and any structures that may lead to further restrictions must not be damaged. “Aphasia” is a term for speech impairment after surgery, for example. “We want accurate knowledge of the risk of aphasia before the operation.”
The chief physician at the Clinic for Neurosurgery at the Klinikum Rechts der Isar has been studying preoperative brain mapping for more than 10 years. “We have long known the basic locations in the brain that are responsible for functions such as movement or speech. But it has only been in the past five years or so that we have begun to analyze the brain network to see how different regions work together, for example to enable a person to speak. There is something One is clear: there is no language center as such. Instead, the structure is more like several hubs or nodes of a large network through which speech can be made.
Brain tumor: making predictions through machine learning
Analysis of brain network properties – referred to as neural network analysis – a process Professor Craig’s team has been using for almost two years – plays a key role in the current research. “This way we determine the number of connections in individual brain regions,” says Professor Craig. “We have since begun to assign functions more precisely to regions of the brain.”
TUM scientists Dr. Haosu Zhang and Dr. Sebastian Ille have mapped anatomical layers of the brain responsible for speech capabilities. The process is as follows: “With a special form of MRI known as tractography, we produce three-dimensional representations of the networks and subnetworks of neural pathways in the brain,” Chang explains.
This network analysis is supported by the process of transcranial magnetic stimulation, in which a targeted magnetic pulse inhibits neurons in the fiber pathways responsible for speech. This causes a temporary speech impairment in the patient that is identifiable in the video analysis. It enables researchers to pinpoint the regions of the brain responsible for speech.
We combine so-called neural network parameters from imaging of the nervous system with information regarding a patient’s speech function.”
Dr. Haosu Zhang, TUM Scientist
What makes Zhang and Ille’s algorithm special: It produces “statistically significant parameters” – data that can be used to train a machine learning model and thus to locate the speech of individual patients. It may seem as complicated as using different methods of analysis – the defining characteristic of the method is its simplicity: the entire analytical process works without complex algorithms or powerful computers. “The data we use is from routine hospital exams,” says Zhang.
Network analysis: 73% accuracy in predicting speech impairment
In a recent study of 60 patients, researchers at the Klinikum rechts der Isar showed that this composite analysis could predict with great accuracy (73%) whether surgery would cause speech difficulties (postoperative aphasia). “It’s very important to be able to make these predictions,” Craig says. He’s excited about the possibility of being able to identify risk more accurately by means of “real network analysis” and having concrete data to support his brain mapping.
Moreover: With the help of machine learning, predictions will get better over time. But for this, researchers will need more patient data to train machine learning algorithms. “It’s the only approach that can use big data to predict the risks of surgery,” says Professor Craig, who now plans to find more patients to take part in his research. He believes that even “a few hundred” patients would be sufficient for highly accurate predictions.