- Catching cancer in a different light
Chemometrics helps analyze terahertz spectra
Investigators at Riken Institute in Sendai and at the University of Fukui, both in Japan, have demonstrated that terahertz imaging, when combined with the correct analysis techniques, could be useful as a cancer diagnostic. In three of four test samples, a terahertz imaging-based approach successfully differentiated cancerous from normal tissue.
In theory, imaging with terahertz wavelengths (100 to 1000 μm) offers significant advantages for medical diagnostics. It is nonionizing, and it deeply penetrates tissue. Therefore, it offers the ability to completely image tissue samples 1.0 mm thick without damage.
The problem, according to Hiromichi Hoshina of Riken, lies in understanding the terahertz imaging measurements. In addition to molecular vibrations, the spectrum also is affected by intermolecular vibrations, powder scattering and optical interference in the sample. Even something like cracks in a sample can affect the terahertz spectra acquired.
Researchers used terahertz images to detect cancer. In the photographs (left) of four cancerous tissue samples (a-d), the cancer areas are marked by red lines. The second and third columns show maps of the absorbance and the refractive index at 1.7 THz, respectively, to the same scale as the photographs. The color-coded absorbance ranges from 0 to 2.0 for samples a, b and d, and from 0 to 0.7 for sample c. The refractive index is shown in color from 1.0 to 1.7 THz for all samples. Images reprinted with permission of Applied Physics Letters.
Interpreting such a complex combination of features is challenging. While dealing with this problem for another research project, Hoshina realized that the terahertz spectra looked similar to those of the near-infrared. In that region, chemometrics — the application of mathematical or statistical methods to chemical data — is a popular and well-suited spectral analysis technique. The group, therefore, applied two chemometrics techniques to the terahertz imaging problem: principal component analysis and hierarchical clustering analysis.
Principal component analysis simplifies data by reducing it to principal components, translating the raw data into a multidimensional set of coordinates. Once the translation is done, most of the variation in the data can be accounted for with only a handful of the principal components. Hierarchical clustering analysis uncovers groupings by looking at the distance between points in principal-component space. The assumption is that points that lie close together have similar properties and, therefore, should be in one group, while those that lie far apart are dissimilar and should be separated.
In their studies, the researchers used a custom time domain terahertz spectroscopy instrument, on which they collaborated with Oø tawara-based Tochigi Nikon Co., which is part of Nikon Corp. of Japan. “We decided on the spectrometer specifications, and they built it,” Hoshina said.
The researchers collected transmission multispectral terahertz images of samples. In a proof of principle, they used it to measure a sample’s absorbance and refractive index spectra from 0.5 to 2.25 THz with 25-GHz resolution. For every 250-μm2 pixel spot, they measured and averaged 40 spectra.
When analyzing the terahertz image data, the researchers used principal component analysis. Here are score plots of the first (PC1) and the second (PC2) principal components. Each column shows the result of principal component analysis relative to the absorbance spectra, to the refractive index spectra and to both. The pixels of the cancer tumor are shown in red.
The investigators used four slices of unstained paraffin-embedded liver cancer tissue as samples, in part because water absorbs quite strongly in the terahertz region, making it difficult to measure the spectrum of a fresh sample.
As detailed in the Jan. 22 issue of Applied Physics Letters, they measured the spectra from samples in which cancerous tissue had been identified by optical microscopy. In comparing the outline of the known cancer with what the terahertz device measured, they found that absorbance could highlight a diseased region. The same was true of the refractive index. However, in some cases, the absorbance increased in the cancer area and decreased in others. A similar variability was seen regarding the refractive index.
The scientists performed principal component analysis on the merged absorbance and refractive index data and followed that with clustering analysis. Together these showed that a systematic diagnosis of the diseased area was possible — in three of the four samples. The fourth, in contrast, did not have a cancer that could be distinguished from normal tissue using terahertz imaging.
This discrepancy might have resulted from viable cancer cells being present in the first three samples, while the fourth had a considerable amount of necrotic cells instead. Differences in fine tissue structure might account for the inconsistency as well. Additional information is needed to prove or disprove the source of the difference, Hoshina noted.
Even with that unresolved issue, he said that terahertz imaging offered several advantages over optical microscopy and could, therefore, be a potentially useful complement to the latter. It can diagnose samples automatically, unlike optical microscopy, where well-trained personnel are needed. The terahertz device can scan an area 4 cm on a side, much larger than the relatively small area covered by an optical microscope. In fact, the wider terahertz scan turned up an unexpected cancerous area that the visual inspection had missed during the proof of concept.
Finally, terahertz imaging can be achieved in hours instead of days because the technique does not require samples to be stained. In the future, that time might be reduced even more. “We are planning to measure frozen samples because ice is more transparent in (the) terahertz region than water. If this is possible, the sample preparation will be faster,” Hoshina said.
MORE FROM PHOTONICS MEDIA