A clearly defined border differentiates benign tumors from malignant ones. Malignant tumors start to get fuzzy around the edges as they begin infiltrating the surrounding healthy tissue. At this later stage of tumor progression, instead of surgical removal, oncologists need to resort to much more aggressive therapeutic interventions to manage the disease, with the risks to patients spiking dramatically. Because of this, early tumor detection is paramount.
The problem is that it’s sometimes difficult to “see” tumors in order to accurately diagnose and gauge their status. Tumors growing deep within internal organs can be inaccessible for biopsies, and because they are surrounded by mucosal capsules, traditional diagnostic imaging practices such as endoscopies don’t always cut it. Cancer experts consider gastrointestinal stromal tumors, or GISTs, among the most notoriously difficult to diagnose—they’re tough to spot and take a long time to assess clinically.
Japanese researchers have come forward with a solution for overcoming the challenges of diagnosing GISTs with a novel technology that combines near-infrared hyperspectral imaging (NIR-HSI) with machine learning.
"This technique is a bit like X-rays, the idea is that you use electromagnetic radiation that can pass through the body to generate images of structures inside," said Hiroshi Takemura who led the development of the new tech.
"The difference is that X-rays are at 0.01-10 nm, but near-infrared is at around 800-2500 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."
This development, published in Nature's Scientific Reports, is the result of a collaborative project between engineers and doctors for exploring the use of NIR-HSI for deep tumor imaging applications.
The creation of the new platform was a three-step process. First, excised tumors from 12 patients positively diagnosed with GISTs were obtained and imaged using NIR-HSI. Next, pathologists examined and analyzed the images captured, delineating the differences between healthy and malignant tissues. This information was then used as training material for a machine-learning algorithm to help automate the pathologists’ diagnoses.
The results were promising, with the algorithm successfully able to color code tumor vs non-tumor tissues with 86 percent accuracy.
"This is a very exciting development," explained Takemura.
"Being able to accurately, quickly, and non-invasively diagnose different types of submucosal tumors without biopsies, a procedure that requires surgery, is much easier on both the patient and the physicians."
There is still a way to go before full clinical implementation of the technology, although the researchers are confident that follow-up studies will help support this goal. Their strategies include expanding their machine learning training dataset to include other types of tumors and integrating the platform with existing endoscopy technologies.