A novel approach combines advanced imaging with artificial intelligence to offer real-time tumor detection.
During cancer surgery, surgeons sometimes extract tissue samples for lab analysis. This is an important step that allows medics to perform more accurate diagnoses and direct the course of treatment, which may include a subsequent surgery to remove the tumor.
The new study compared the ability of an AI to detect tumors in these samples with the ability of competent pathologists. The AI-based diagnosis software was 94.6% accurate, compared to 93.9% for the pathologist interpretation. It also works in near-real-time, with the diagnosis taking little over 2 minutes.
Over 1 million brain samples are analyzed in the US alone every year, a process that is time-, resource-, and labor-intensive. To add even more to this problem, vacancies in neurology departments are not uncommon.
With this in mind, neuroscientists Daniel Orringer and his colleagues set out to develop a new diagnostic tool. It combines a powerful optical imaging technique, called stimulated Raman histology (SRH), with an artificially intelligent deep neural network. During surgery, images are acquired through SRH and then fed to the AI algorithm, which makes the assessment in 150 seconds.
Pathologists are generally accurate, but this approach can greatly reduce the time and effort needed for diagnosis. As an added bonus, the AI is also capable of detecting features that can escape the human eye.
“As surgeons, we’re limited to acting on what we can see; this technology allows us to see what would otherwise be invisible, to improve speed and accuracy in the [operating room], and reduce the risk of misdiagnosis,” Orringer, the senior author of the paper, said in a press statement. “With this imaging technology, cancer operations are safer and more effective than ever before.”
The researchers trained the AI using more than 2.5 million samples, classifying them in different categories that represent the most common types of brain tumors. The algorithm was then tested for efficiency on 278 brain tumor and epilepsy patients, and its results compared to that of human doctors. Neither the AI nor the pathologists are perfect but there’s an upside to this: the errors that the AI did were different from the ones that humans made. This suggests that, should a pathologist and an AI analyze the same tissue sample, they might come very close to 100% accuracy. This means that the AI could be used both to complement the lack of neuroscientists or to complement them and improve the results.
Slowly but surely, AI is starting to enter the medical room — and it can make a real difference.
The study has been published in Nature Medicine.
Was this helpful?