According to researchers, the technology may eventually be used to treat different types of cancer.
In only ten seconds, researchers have created an AI-powered model that can identify whether any portion of a malignant brain tumour that may be excised is still present during surgery.
The tool, known as FastGlioma, performed significantly better than traditional techniques for determining what is left of a tumour. According to researchers, it could revolutionise neurosurgery by instantly enhancing the overall care of patients with diffuse gliomas.
The technology may be used to additional paediatric and adult brain tumour diagnosis and is quicker and more accurate than the current standard of care approaches for tumour detection. It might be useful for
In depth:
It might be used as a basis for directing surgery for brain tumours. Rarely can a neurosurgeon remove the entire mass when removing a potentially fatal tumour from a patient’s brain. Residual tumour is what’s left over.
Because surgeons cannot distinguish between a healthy brain and a tumour that remains in the cavity where the mass was removed, the tumour is frequently overlooked during the procedure.
One of the biggest surgical challenges is that the remaining tumour may seem like healthy brain.
During a procedure, neurosurgical teams use a variety of techniques to find that remaining tumour.
They might undergo MRI imaging, which necessitates intraoperative equipment that isn’t always accessible. They might undergo MRI imaging, which necessitates intraoperative equipment that isn’t always accessible.
A fluorescent imaging agent may also be used by the surgeon to detect tumour tissue, albeit this isn’t always the case. Their broad use is hindered by these restrictions.
The research:
Neurosurgical teams examined new, unprocessed specimens taken from 220 patients who underwent surgery for low- or high-grade diffuse gliomas as part of this global research utilising AI-driven technology.
FastGlioma had an average accuracy of about 92% in detecting and estimating the amount of tumour that was left. When comparing surgeries guided by image- and fluorescence-guided techniques or FastGlioma forecasts, the AI system missed high-risk, residual tumours only 3.8% of the time, while conventional methods missed them over 25% of the time. Researchers used 4 million distinct microscopic fields of view and more than 11,000 surgical specimens to pre-train the visual foundation model before creating FastGlioma.
Stimulated Raman histology, a fast, high-resolution optical imaging technique created at U-M, is used to image the tumour samples. DeepGlioma, an AI-based diagnostic screening system that can identify genetic alterations in brain tumours in less than 90 seconds, was trained using the same methodology.
Stimulated Raman histology takes about 100 seconds to collect full resolution images, while a “fast mode” lesser resolution image only takes 10 seconds. The full resolution model’s accuracy reached up to 92%, while the rapid mode’s accuracy was marginally lower at about 90%, according to the researchers.
Take away:
In addition to being a convenient and reasonably priced tool for neurosurgeon teams working on gliomas, researchers claim that FastGlioma may precisely identify residual tumour for a number of non-glioma tumour diagnoses’, such as meningiomas and paediatric brain tumours including medulloblastoma and ependymoma.
According to co-author Aditya S. Pandey, M.D., chair of the Department of Neurosurgery at U-M Health, these findings show the value of visual foundation models like FastGlioma for medical AI applications and the possibility of generalising to other human cancers without the need for significant model retraining or fine-tuning.
New technology, such as sophisticated imaging and artificial intelligence techniques, have been suggested to be included into cancer surgery by global cancer efforts.
