A research team led by University of Michigan and University of California San Francisco has developed an AI powered model that, in just 10 seconds, can determine during surgery if any part of a cancerous brain tumor that could be removed remains.
This revolutionary technology, named FastGlioma, has demonstrated a remarkable advantage over traditional methods for tumor identification, as highlighted by the expert research team from the University of Michigan and the University of California, San Francisco.
“FastGlioma is an artificial intelligence-based diagnostic system that has the potential to change the field of neurosurgery by immediately improving comprehensive management of patients with diffuse gliomas,” said senior author Todd Hollon, M.D., a neurosurgeon at the University of Michigan Health and assistant professor of neurosurgery at U-M Medical School.
“The technology works faster and more accurately than the current standard of care methods for tumor detection and could be generalized to other pediatric and adult brain tumor diagnoses. It could serve as a foundational model for guiding brain tumor surgery.”
When a neurosurgeon tackles a life-threatening tumor in a patient’s brain, they are rarely able to remove the entire mass. The leftover tissue, referred to as the residual tumor, poses significant risks.
During surgery, surgeons frequently struggle to distinguish between healthy brain tissue and residual tumors within the cavity. Residual tumors can closely resemble healthy brain tissue, posing a significant challenge during operations.
Neurosurgical teams utilize various techniques to identify residual tumors during a surgical procedure. They might obtain MRI imaging, which necessitates specialized machinery that isn’t available everywhere.
Additionally, fluorescent imaging agents can highlight tumor tissues, but their effectiveness is limited to certain tumor types. These constraints highlight the urgent need for advancements in surgical methods to ensure comprehensive tumor removal and enhance patient safety.
In an international study utilizing advanced AI technology, neurosurgical teams examined fresh, unprocessed specimens from 220 patients who underwent surgery for low- and high-grade diffuse gliomas. FastGlioma demonstrated an impressive average accuracy of approximately 92% in detecting and quantifying remaining tumor tissue.
When comparing surgeries guided by FastGlioma’s predictions to those using traditional image- and fluorescent-guided techniques, this innovative AI solution showed a remarkable 3.8% miss rate for high-risk residual tumors, vastly outperforming conventional methods, which experienced nearly a 25% miss rate.
“This model is an innovative departure from existing surgical techniques by rapidly identifying tumor infiltration at microscopic resolution using AI, greatly reducing the risk of missing residual tumor in the area where a glioma is resected,” said co-senior author Shawn Hervey-Jumper, M.D., professor of neurosurgery at the University of California San Francisco and a former neurosurgery resident at U-M Health. “The development of FastGlioma can minimize the reliance on radiographic imaging, contrast enhancement, or fluorescent labels to achieve maximal tumor removal.”
FastGlioma combines microscopic optical imaging with a form of artificial intelligence known as foundation models to evaluate the remnants of a brain tumor. These sophisticated AI models, like GPT-4 and DALL·E 3, are trained on extensive and varied datasets, allowing them to adapt to numerous applications with remarkable efficiency.
After undergoing large-scale training, foundation models can perform tasks such as image classification, serve as chatbots, answer emails, and create images based on text prompts.
In developing FastGlioma, researchers pre-trained the visual foundation model using more than 11,000 surgical samples and 4 million distinct microscopic views. Tumor specimens are analyzed through stimulated Raman histology, a rapid and high-resolution optical imaging technique established at U-M. This same technology was utilized to develop DeepGlioma, an AI diagnostic screening tool capable of identifying genetic mutations in brain tumors in less than 90 seconds.
“FastGlioma can detect residual tumor tissue without relying on time-consuming histology procedures and large, labeled datasets in medical AI, which are scarce,” said Honglak Lee, Ph.D., co-author and professor of computer science and engineering at U-M.
High-resolution images can be captured in approximately 100 seconds using stimulated Raman histology, while a “fast mode” that offers lower resolution images can be obtained in just 10 seconds. The research demonstrated that the full resolution approach reached an accuracy of up to 92%, whereas the fast mode was slightly lower at around 90%.
“This means that we can detect tumor infiltration in seconds with extremely high accuracy, which could inform surgeons if more resection is needed during an operation,” Hollon said.
Despite advancements over the past two decades, the rates of residual tumors following neurosurgery remain stagnant. Residual tumor presence not only leads to deteriorating quality of life and premature deaths for patients, but it also places an increasing strain on healthcare systems projected to manage 45 million surgical procedures globally by 2030.
Recognizing the urgency, global cancer initiatives advocate for the integration of innovative technologies, such as advanced imaging techniques and artificial intelligence, into cancer surgical practices. The 2015 Lancet Oncology Commission on Global Cancer Surgery underscored the critical need for cost-effective strategies to tackle surgical margins, driving the demand for breakthrough technologies.
FastGlioma stands out as a highly accessible and affordable resource for neurosurgical teams tackling gliomas; furthermore, research indicates it can accurately identify residual tumors in various non-glioma diagnoses, including pediatric brain tumors like medulloblastoma and ependymoma, as well as meningiomas.
“These results demonstrate the advantage of visual foundation models such as FastGlioma for medical AI applications and the potential to generalize to other human cancers without requiring extensive model retraining or fine-tuning,” said co-author Aditya S. Pandey, M.D., chair of the Department of Neurosurgery at U-M Health. “In future studies, we will focus on applying the FastGlioma workflow to other cancers, including lung, prostate, breast, and head and neck cancers.”
Journal reference:
- Akhil Kondepudi, Melike Pekmezci, Xinhai Hou, Katie Scotford, Cheng Jiang, Akshay Rao, Edward S. Harake, Asadur Chowdury, Wajd Al-Holou, Lin Wang, Aditya Pandey, Pedro R. Lowenstein, Maria G. Castro, Lisa Irina Koerner, Thomas Roetzer-Pejrimovsky, Georg Widhalm, Sandra Camelo-Piragua, Misha Movahed-Ezazi, Daniel A. Orringer, Honglak Lee, Christian Freudiger, Mitchel Berger, Shawn Hervey-Jumper & Todd Hollon. Foundation models for fast, label-free detection of glioma infiltration. Nature, 2024; DOI: 10.1038/s41586-024-08169-3