Advancing in vivo terahertz imaging

Terahertz (10¹²Hz, THz) frequency radiation lies between the IR and millimeter regions of the electromagnetic spectrum (1THz corresponds to 33.3cm⁻¹, 4.14meV, and has a wavelength of 300μm). Historically, terahertz wave generation was a challenge. Various methods to extend optical techniques to lower frequencies or to ramp up electronic techniques to higher frequencies have been explored. Significant advances in the field mean that presently, numerous companies sell systems and components worldwide (see Figure 1). As generating terahertz radiation is no longer a hurdle, the interest now lies in finding potential applications for this radiation.

Figure 1. (a) Photograph of our terahertz imaging probe system from TeraView Ltd, UK. (b) The imaging window is made of quartz and is 1.5cm long.

From a safety perspective, terahertz imaging would be an ideal modality for patient screening and diagnosis. In most systems that use a photoconductive antenna, the average power is less than 1μW. This is a million times lower than the radiation naturally emitted by humans (1W). Additionally, the frequency is much lower than what is used for x-ray imaging (10²⁰Hz), and, unlike x-ray radiation, its frequency is too low to cause ionization. Thus, terahertz imaging poses no known safety risk to humans at the power levels that have been measured in photoconductive systems. It is a non-destructive, non-ionizing imaging modality.

Inter-molecular forces, such as hydrogen bonds, have resonances extending into the terahertz region. This means that the penetrative depth of terahertz radiation into biological tissue with high water content is very shallow—of the order of millimeters depending on tissue type. As a result, in vivo tissue imaging requires reflection geometry. However, this heightened sensitivity to hydrogen bonding and other intermolecular interactions is an advantage. Subtle changes in the water content and structure of tissues can be relevant when identifying abnormalities such as cancer. For example, terahertz images of freshly excised breast cancer (of the in situ non-calcified form) have shown contrast.¹ This is particularly significant because non-calcified tumors are often missed during breast-conserving surgery because they are not visible by x-ray imaging nor are they palpable.

In order to understand and investigate contrast mechanisms, we measured the terahertz spectroscopic properties of skin cancer,² breast cancer,³ and liver cirrhosis.⁴ For all of these aberrations, we found that at terahertz frequencies the abnormal tissue had higher absorption coefficients than the corresponding healthy tissue. For liver tissue we also devised a study to quantify the extent of the change in absorbance coefficient due to changes both in water content and structure. We found that for frequencies above 0.4THz, structural changes contributed to 50–66% of the total change in absorption coefficient. Thus, water was neither the sole nor dominant source of contrast in the terahertz properties of liver cirrhosis,⁴ as previously thought.

We also conducted in vivo measurements of human skin. Of the three top layers of the skin, some interfaces cause reflection: see Figure 2(a). Our handheld terahertz probe has a more flexible geometry than our previous flat-bed system, so it can be waved over the skin easily: see Figure 2(b). Despite many advantages, including ease of use, the terahertz signal is noisier and of lower bandwidth for this probe. Thus, we developed a more effective method to remove unwanted reflections from the lower surface of the quartz window.⁵ The ‘baseline’ (unwanted reflections) is subtracted from the raw data, and the resultant data is processed further (namely by de-convoluting a reference pulse and applying a bandpass filter) to obtain the waveforms shown in Figure 2(c). Using the new baseline, we obtained a waveform that is consistent with our previous skin study using the flat-bed imaging system⁶ but without the artifacts previously encountered on either side of the main pulse. We also devised a frequency and wavelet domain de-convolution (FWDD) process to better filter the terahertz system response (see Figure 3).⁷ This enabled the resolution of thinner features than if a bandpass filter was used.

Figure 2. (a) Schematic diagram showing the layers of skin against the quartz window of the probe. (b) Photograph of the probe being used to measure the skin on the palm. (c) The resulting reflected waveforms after processing using the standard (left) and new (right) baseline methods. The separation of the two reflected troughs is related to the thickness of the stratum corneum.

Figure 3. Resulting waveforms using the new baseline method and applying (a) band pass filter after de-convolution, and (b) frequency-wavelet domain de-convolution (FWDD).

Our new processing methods have enabled us to use the probe to resolve features to similar levels as with the flat-bed system. We are now poised to conduct further in vivo investigations.