Brain surgeons often grapple with critical decisions throughout the tumor removal process.
Now, Dutch scientists have introduced an AI tool to aid the intra-surgical decision-making process, providing surgeons with ultra-fast insights into tumor types and subtypes.
According to a recent study published in Nature, the AI examines specific segments of a tumor’s DNA, identifying distinct chemical structures.
This analysis provides insights into the tumor’s type and possibly its subtype. Such timely information can guide surgeons in their approach to the surgical operation.
“It’s imperative that the tumor subtype is known at the time of surgery,” stated Jeroen de Ridder, an associate professor at UMC Utrecht. “What we have now uniquely enabled is to allow this very fine-grained, robust, detailed diagnosis to be performed already during the surgery.”
Their deep learning system, named Sturgeon, was subject to rigorous testing. In certain initial tests, the AI refrained from diagnosing due to ambiguous data.
Overall, the researchers demonstrated the model’s real-time effectiveness across 25 surgeries, achieving a rapid diagnostic turnaround time of less than 90 min, much quicker than traditional methods. 72% of diagnoses were correct, but seven didn’t reach the required confidence threshold.
While the standard diagnostic process involves a microscopic examination of brain tumor samples, comprehensive genetic sequencing offers deeper insights. However, as Dr. Alan Cohen, from Johns Hopkins highlighted, “We have to start treatment without knowing what we’re treating.”
Dr. de Ridder further explained the AI’s capability: “It can figure out itself what it’s looking at and make a robust classification.”
Still, some challenges persist. Variability within the tumor, sample size, and certain elusive tumors can pose difficulties. On this, Marc Pagès-Gallego, a study’s co-author, provided some context on how they navigated these issues.
Dr. Sebastian Brandner from University College London commented on the practicalities, asserting, “Implementation itself is less straightforward than often suggested.”
While the tool represents a significant stride forward, it does have its limitations.
As Dr. Cohen admitted, “We’ve made some gains, but not as many in the treatment as in the understanding of the molecular profile of the tumors.”
More about the study
The innovative tool ‘Sturgeon’ uses deep learning to enhance the intraoperative classification of brain tumors, aiding in better surgical decision-making.
The model is designed to be ‘ultra-fast’ in classifying tumor types and subtypes.
- Challenges: Classifying tumors during surgery is tough due to limited sequencing time and uncertain data coverage.
- Sturgeon’s solution: This deep learning system uses data derived from widely available methylation array data, which is used to profile brain tumors. Sturgeon outsources intensive computational tasks to minimize computational resources during surgery.
- Performance: The model has shown consistent results but often doesn’t account for intratumor diversity. It provided the correct diagnosis in 72% of surgical tests.
- Future development: As more data is acquired, Sturgeon’s data will be enhanced, though data-sharing restrictions due to privacy concerns make cross-institutional learning tricky.
- Limitations: One potential limitation is the tissue amount required. The model requires a sample size of about 5 x 5 x 5 mm for best results, but sufficient DNA has been successfully extracted from smaller samples.
Crucially, with Sturgeon, tumor diagnostics results can be achieved within 90 minutes, aligning with surgical timelines. This empowers surgeons with real-time insights, enabling them to make better decisions during surgery.
While Sturgeon’s data is beneficial, it would ideally be used alongside a trained pathologist’s assessment, according to the study.
Sturgeon joins legions of cutting-edge medically-oriented AI models created this year, illustrating the technology’s ability to transform healthcare outcomes.
