https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html https://www.euro-online.org/enog/inoc2007/Papers/m https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html

Early cancer diagnosis with NIR spectroscopy

As is the case with any cancer, early diagnosis of colorectal cancer, which affects the colon, rectum and appendix, greatly increases the chance of successful treatment. At the moment, however, most colorectal cancer is only diagnosed after a patient has become ill, by which time the cancer is fairly well advanced and has often spread.

Potentially, colorectal cancer could be diagnosed much earlier, because cancerous tissue appears well before any symptoms and this tissue has a different chemical composition to healthy tissue. For example, cancerous tissue generally has lower levels of carbohydrates and phosphates than healthy tissue. These kinds of differences in chemical composition can be picked up by NIR spectroscopy, based on the different response profiles of biomolecules such as lipids, carbohydrates and proteins.

To explore whether these responses would be sufficiently sensitive to diagnose colorectal cancer, Chinese researchers led by Chao Tan at Yibin University used NIR spectroscopy to study healthy and cancerous tissue extracted from 20 colorectal cancer patients. Because they studied several areas of each tissue sample, comprising healthy and cancerous cells, this produced 186 different spectra.

After removing background noise, Tan and his colleagues analysed the 186 spectra with principal component analysis (PCA). This showed that the healthy and diseased samples could be divided into two broad groups, with just three principal components accounting for 96% of the variance between the different samples. However, the two groups did overlap somewhat and so Tan and his colleagues tested several different mathematical techniques for producing predictive models from a subset of this spectra. These comprised random forest (RF), partial least squares-discriminant analysis, K-nearest neighbour, and classification and regression tree.

As they report in Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, RF produced the most accurate model. When tested on the subset of the spectra not used to produce it, the RF model correctly classified all the healthy samples and all but four of the 39 cancerous samples. What is more, Tan and his colleagues uncovered evidence that NIR spectroscopy can not only distinguish cancerous tissue from healthy tissue, but can also determine the stage of the cancer. The next step, says Tan, is to test this technique on many more samples of cancerous tissue, although ideally you would want to take advantage of NIR spectroscopy’s non-invasiveness to identify cancerous tissue still in the body.

Blog tags: 
Techniques: