Donate to Science & Enterprise

S&E on Mastodon

S&E on LinkedIn

S&E on Flipboard

Please share Science & Enterprise

Laser Images, A.I. Diagnose Brain Tumors

Brain tumor graphic

(National Science Foundation)

6 Jan. 2020. Results of a clinical trial show laser images analyzed by machine learning algorithms can offer near real-time diagnosis of brain tumor tissue during surgery. A description of the technology and findings from the trial, conducted by researchers at New York University and University of Michigan, appear in today’s issue of the journal Nature Medicine (paid subscription required).

A team led by neurosurgery professor Daniel Orringer, on the faculty at New York University and University of Michigan medical schools, is seeking faster detection of brain tumor tissue during surgery to remove the tumors. In most cases today, the process requires removal of brain tissue from the patient, then prepared, stained, and reviewed by a pathologist in a separate lab. Assuming the pathologist is immediately available for the task, the diagnosis can take 20 to 30 minutes, while the surgical team waits for the results.

The new technology uses brain tissue biopsies, similar to current methods. But the new process, called stimulated Raman histology, or SRH, requires only a single preparation step, followed by exposure to near-infrared laser beams that capture microscopic images at up to 30 frames per second. The lasers excite molecules that vibrate and scatter the light beams in a process called Raman spectroscopy, with vibrations then amplified using a second light source. The different vibrations emitting from molecules in the tissue make it possible to identify and distinguish between various types of cells, including tumor cells.

To quickly interpret the images, the researchers developed algorithms using a convolutional neural network. These algorithms combine image analysis and machine learning to dissect an image by layers for understanding features in the image. Different aspects of each layer discovered and analyzed by the system are translated into data that the algorithm then uses to train its understanding of the problem being solved, with that understanding enhanced and refined as more images and data are encountered.

The team reports using more than 2.5 million SRH images from 415 patients to train its algorithms that group brain tissue image samples into 13 major classes of brain tumors. The algorithms also identify key diagnostic regions within tumor images. The authors say the system returns results in under 2.5 minutes.

Orringer and colleagues at three university medical centers recruited 278 patients undergoing brain surgery for tumors or epilepsy to assess the technology. Biopsies from the patients were randomly assigned for conventional review in pathology labs or the SRH process, with both techniques returning almost equally accurate results: 94 percent for conventional pathology and 95 percent for SRH. The SRH results, however, were returned to surgeons in near real-time.

“As surgeons, we’re limited to acting on what we can see,” says Orringer in an NYU statement. “This technology allows us to see what would otherwise be invisible, to improve speed and accuracy in the OR, and reduce the risk of misdiagnosis.”

The SRH technology is based on research originally conducted at Harvard University and licensed for commercial development to Invenio Imaging Inc., a company in Santa Clara, California. Orringer is a scientific adviser to Invenio Imaging.

More from Science & Enterprise:

*     *     *

1 comment to Laser Images, A.I. Diagnose Brain Tumors