Novel AI-based algorithm accurately diagnoses brain tumors
Researchers have leveraged advanced optical imaging techniques and an artificial intelligence (AI)-based algorithm to develop a novel imaging system that accurately detects brain tumors.
Researchers from the University of Michigan (MI, USA) and New York University’s School of Medicine and Langone Health (both NY, USA) have leveraged advanced optical imaging techniques and an artificial intelligence (AI)-based algorithm to develop a novel imaging system that provides accurate, real-time brain tumor diagnoses.
In the study, published today in Nature Medicine, investigators compared the diagnostic accuracy of the newly developed imaging system – which leverages stimulated Raman histology imaging and AI – with that of a group of trained pathologists supplied with conventional brain histologic images.
Stimulated Raman histology collects scattered laser light to detect tumor infiltration in human tissue by revealing and illuminating image features typically overlooked in standard histologic image examinations.
Researchers applied an AI algorithm – based on a deep, convolutional neural network trained with over 2.5 million histologic samples from 415 patients – to images scrutinized by stimulated Raman histology imaging to predict brain tumor diagnoses in under 3 minutes. Further, following surgical resection of suspected tumors, the same technology may allow accurate identification, and subsequent removal, of otherwise undetectable tumor cells.
The images used to train the AI-based algorithm encompassed 13 histologic categories, collectively representing the most common brain tumors, including malignant glioma, lymphoma, metastatic tumors and meningioma.
To evaluate the comparative diagnostic accuracy of the new imaging technology versus the standard-of-care brain tumor diagnostic tool – pathologist diagnosis – researchers conducted a clinical trial involving 278 patients undergoing tumor resection across three university medical centers. Investigators split resected brain tumor specimens into sister samples that were randomly assigned for scrutiny by the new imaging technology, or conventional pathologists.
The diagnostic accuracy of the AI-based technology was observed to be 94.6%; this was statistically similar to the accuracy of pathologists’ interpretations of histologic images, which was 93.9%.
Importantly, the diagnostic errors observed in the experimental and control trial groups were distinct; this suggests that experienced pathologists who leverage the new imaging system could achieve close to 100% brain tumor diagnostic accuracy.
Senior study author Daniel Orringer (New York University’s Grossman School of Medicine) explained: “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 concluded: “With this imaging technology, cancer operations are safer and more effective than ever before.”
Hollon TC, Pandian B, Adapa AR et al. Near real-time intraoperative brain tumor diagnosis using stimulated Raman histology and deep neural networks. Nat. Med. doi:10.1038/s41591-019-0715-9 (2020) (Epub ahead of print);