Machine Learning-Enabled NIR Hyperspectral Imaging System IDs Hidden Tumors

Collaborating researchers from Tokyo University of Science, National Cancer Center Hospital East, and RIKEN Center for Advanced Photonics have developed a technology using near-infrared hyperspectral imaging (NIR-HIS) and machine learning that finds hidden tumors, such as those in deep tissue and/or covered by a mucosal layer.

These tumors are difficult to find with standard methods such as endoscopy. Gastrointestinal stromal tumors (GISTs), often found in the stomach and small intestines, for example, can require time-consuming and difficult techniques that can prolong diagnosis.

The new technique aims to improve that process specifically. Hiroshi Takemura, a doctor at Tokyo University of Science, said that while the method is a bit like x-raying, it relies on working with near-infrared light with wavelengths of around 800 to 2500 nm.

“At that wavelength, near-infrared radiation makes tissues seem transparent in images. And these wavelengths are less harmful to the patient than even visible rays,” Takemura said.

The team tested the technique by imaging 12 patients with confirmed cases of GISTs who had their tumors surgically removed. The researchers imaged the excised tissues using NIR-HIS, and a pathologist examined the images to determine the border between normal and tumor tissue. The images were then used to train data for a machine-learning algorithm to allow it to distinguish between pixels representing healthy and tumor tissue.

The researchers found that even with 10 of the 12 tumors covered with a mucosal layer, the algorithm was able to identify GISTs and correctly color code tumor and nontumor sections with accuracy greater than 85%.

The researchers intend to expand the training data set, improving accuracy and allowing the team to add information about the depth of the tumor and about other types of tumors. Work is currently underway to develop an NIR-HIS system that builds on existing endoscopy technology.

“We’ve already built a device that attaches an NIR-HIS camera to the end of an endoscope and hope to perform NIR-HIS analysis directly on a patient soon, instead of just on tissues that had been surgically removed,” Takemura said. “In the future, this will help us separate GISTs from other types of submucosal tumors that could be even more malignant and dangerous. This study is the first step toward much more groundbreaking research in the future, enabled by this interdisciplinary collaboration.”

The research was published in Scientific Reports (www.doi.org/10.1038/s41598-020-79021-7).