Diffuse gliomas are the most common malignant brain tumors in adults: from the outset they infiltrate healthy tissue and invariably progress toward malignant transformation. Surgery, however, prolongs survival despite their "invisibility" if as much tumor volume as possible can be safely removed. New intraoperative imaging and the first experiments with artificial intelligence promise help, but they also bring limitations and risks.
Why diffuse gliomas are so hard to treat
On MRI they may appear sharply demarcated, but in reality tumor cells spread centimeters beyond the visible margin and often blend with healthy brain under the microscope. Gliomas also infiltrate functionally critical areas and tracts, for example the fasciculus arcuatus, so some tissue must be preserved to prevent the patient from losing speech or other abilities. The most common initial symptom is a seizure, but manifestations are varied and depend on location.
The legitimate question "why operate" is answered by the data: if the widest possible yet safe resection is achieved, the patient lives longer—this holds even with subtotal resection. This relationship has been known for years and is confirmed by newer, higher-quality studies. The stumbling block remains that part of the tumor is practically indistinguishable from brain by sight and touch during surgery.
Ultrasound in the operating room: a helper with troublesome artifacts
Since intraoperative MRI is not available in Slovakia, neurosurgeons primarily use three-dimensional ultrasound fused with preoperative MRI. It can visualize the extent of the tumor in real time, but has a weakness: at the interface between the resection cavity and fluid, so-called acoustic artifacts arise. These can obscure residual tumor or, conversely, feign a focus where none exists.
A team from Bratislava tested artifact suppression with a microprobe and later joined an international collaboration when Norwegian authors in 2019 demonstrated a special fluid with acoustic properties similar to brain. After filling the cavity with it, the image clears and the true infiltration is more clearly visible. Bratislava was the first center outside Trondheim to use this approach, and it published its experience in specialist journals.
AI during surgery: early results and real risks
These experiences led to a collaboration with Spanish neurosurgeon and programmer Santiago Cepeda. His model processes ultrasound video in real time on a high-performance, isolated laptop and displays probable tumor foci (in red) and the peritumoral area (in green). In pilot cases it identified even small residual tumor and distinguished edema and artifacts; for now, however, the Bratislava center is validating it mainly in terms of its impact on routine workflow, and the model needs further training as well as extension to 3D.
Alongside promising results, the author points to the risks of AI in medicine: hallucinations even in tools for physicians (e.g., contradictions regarding "surgically inaccessible" lesions) and unreliable citations. He therefore implemented a safety protocol for the use of LLMs and also recalls findings from a recent preprint from Cambridge: reliance on LLMs can weaken one's own cognitive effort. AI should be a good servant, not a master—the surgeon bears responsibility, and society must protect the development of children's cognitive abilities as rigorously as it once fought smoking.
