In this work we investigate new feature extraction algorithms on the T-ray response of normal human bone
cells and human osteosarcoma cells. One of the most promising feature extraction methods is the Discrete
Wavelet Transform (DWT). However, the classification accuracy is dependant on the specific wavelet base chosen.
Adaptive wavelets circumvent this problem by gradually adapting to the signal to retain optimum discriminatory
information, while removing redundant information. Using adaptive wavelets, classification accuracy, using a
quadratic Bayesian classifier, of 96.88% is obtained based on 25 features. In addition, the potential of using
rational wavelets rather than the standard dyadic wavelets in classification is explored. The advantage it has
over dyadic wavelets is that it allows a better adaptation of the scale factor according to the signal. An accuracy
of 91.15% is obtained through rational wavelets with 12 coefficients using a Support Vector Machine (SVM) as
the classifier. These results highlight adaptive and rational wavelets as an efficient feature extraction method
and the enormous potential of T-rays in cancer detection.
This study investigates binary and multiple classes of classification via support vector machines (SVMs). A couple of groups of two dimensional features are extracted via frequency orientation components, which result in the effective classification of Terahertz (T-ray) pulses for discrimination of RNA data and various powder samples. For each classification task, a pair of extracted feature vectors from the terahertz signals corresponding to each class is viewed as two coordinates and plotted in the same coordinate system. The current classification method extracts specific features from the Fourier spectrum, without applying an extra feature extractor. This method shows that SVMs can employ conventional feature extraction methods for a T-ray classification task. Moreover, we discuss the challenges faced by this method. A pairwise classification method is applied for the multi-class classification of powder samples. Plots of learning vectors assist in understanding the classification task, which exhibit improved clustering, clear learning margins, and least support vectors. This paper highlights the ability to use a small number of features (2D features) for classification via analyzing the frequency spectrum, which greatly reduces the computation complexity in achieving the preferred classification performance.
We investigate the classification of the T-ray response of normal human bone cells and human osteosarcoma cells, grown in culture. Given the magnitude and phase responses within a reliable spectral range as features for input vectors, a trained support vector machine can correctly classify the two cell types to some extent. Performance of the support vector machine is deteriorated by the curse of dimensionality, resulting from the comparatively large number of features in the input vectors. Feature subset selection methods are used to select only an optimal number of relevant features for inputs. As a result, an improvement in generalization performance is attainable, and the selected frequencies can be used for further describing different mechanisms of the cells, responding to T-rays. We demonstrate a consistent classification accuracy of 89.6%, while the only one fifth of the original features are retained in the data set.
This study investigates the application of one dimensional discrete wavelet transforms in the classification of T-ray pulsed signals. The Fast Fourier Transform (FFT) is used as a feature extraction tool and a Mahalanobis distance classifier is employed for classification. In this work, soft threshold wavelet shrinkage de-noising plays an important part in de-noising and reconstructing T-ray pulsed signals. In addition, Mallat's pyramid algorithm and a local modulus maxima method to reconstruct T-ray signals are investigated. Particularly the local modulus maxima method is analyzed and comparisons are made before and after reconstruction of signals. The results demonstrate that these two methods are especially effective in analyzing and reconstructing T-ray pulsed responses. Moreover, to test wavelet de-noising effectiveness, the accuracy of the classiffication is calculated and results are displayed in the form of scatter-plots. Results show that soft threshold wavelet shrinkage de-noising improves the classification accuracy and successfully generates visually pleasing scatter plots at selected three frequency components.
A simple method to extract the far-infrared dielectric parameters of a homogeneous material from terahertz signals is explored in this paper. Provided with a reference, sample-probing terahertz signal and a known sample thickness, the method can determine the underlying complex refractive index of the sample within a few iterations based on the technique of fixed-point iteration. The iterative process is guaranteed to converge and gives the correct parameters when the material thickness exceeds 200 μm at a frequency of 0.1 THz or 20 μm at a frequency of 1.0 THz.
Terahertz time domain spectroscopy (THz-TDS) has a wide range of
applications from semiconductor diagnostics to biosensing. Recent
attention has focused on bio-applications and several groups have
noted the ability of THz-TDS to differentiate basal cell carcinoma
tissue from healthy dermal tissue ex vivo.
The contrast mechanism is unclear but has been attributed to
increased interstitial water in cancerous tissue. In this work we
investigate the THz response of human osteosarcoma cells and
normal human bone cells grown in culture to isolate the cells'
responses from other effects. A classification algorithms based
on a frequency selection by genetic algorithm is used to attempt
to differentiate between the cell types based on the THz spectra.
Encouraging preliminary results have been obtained.
Pulsed THz imaging systems have a number of potential advantages
in inspection applications. They provide amplitude and phase
information across a broad spectral range in the far-infrared, and
many common packaging materials are relatively transparent in this
frequency range. We use T-ray imaging to allow the identification
of different powdered materials concealed inside envelopes. Using
the terahertz spectral information we show that different powders
may be uniquely identified.
Different thicknesses of the powders are imaged to investigate the
influence of scattering on the measured THz pulses and the
classification model is extended to allow it to identify different
materials independent of the material thickness.
The use of terahertz pulses (T-rays) for imaging has created a wide range of new applications. This paper investigates a number of techniques for optimally classifying terahertz data. Specifically we consider statistical pattern classification methods.
A goal of this research is to implement a classifier as for classifying biomaterials. The objective is to train a classifier using THz images of known materials and then to use the classifier to identify the materials present in unknown images. Potential applications in security systems for airports, customs, and post offices are significant, because we can actually identify the material inside the package without opening it, based on a material’s broadband frequency signature.
We present three T-ray (terahertz wave) tomographic imaging modalities: T-ray computed tomography, T-ray diffraction tomography, and tomographic imaging with a Fresnel binary lens. Each of these techniques uses pulses of broadband terahertz radiation to obtain 3-dimensional images of targets with wide potential application. We present images demonstrating the performance of each technique and discuss their relative advantages.
T-ray imaging and spectroscopy both exploit the terahertz (THz) region of the spectrum. This gives rise to very promising industrial and biomedical applications, where non-invasive and sensitive identification of a substance is achievable, through a material's distinct absorption features in the THz band. Present T-ray systems are limited by low output power, and the race is now on to find more efficient THz emitters. We discuss the feasibility of a novel high-power gallium nitride emitter for terahertz generation. This paper details the advantages of such an emitter, primarily by virtue of its high-voltage capability, and evaluates the benefits of sapphire and silicon carbide substrates. The far-infrared transmission spectra for thin samples of GaN, sapphire and SiC are reported. A high-power THz emitter, that operates at room temperature and is potentially low-cost will open up a host of new possibilities and applications. The central result in this paper demonstrates that sapphire is the better choice over SiC, for the GaN supporting substrate, as we show that it has superior THz transmission characteristics.
There is increasing interest among research groups around the world in the terahertz portion of the electromagnetic spectrum. T-ray systems, driven by ultrafast THz pulses, offer a number of unique advantages over other techniques and are under investigation for a wide range of applications. Biomedical diagnostics is an area of particular emphasis. The sub-millimetre spectroscopic measurements obtained from T-ray systems contain a wealth of information about the sample under test. A number of hurdles, however, hinder the application of T-ray technology. One of the major hurdles to be overcome is the slow acquisition speed of modern THz systems. The chirped probe pulse technique offers a significant improvement in this context. We present results demonstrating the terahertz responses of biological samples measured using a chirped probe pulse, and discuss the problem of data processing and extracting sample characteristics. We show that different types of tissue can be classified based on their terahertz response measured with the chirped probe pulse technique. We consider chicken and beef samples and differentiate between bone and normal tissue. We demonstrate the performance of linear filter models for feature extraction and show that these models are significantly more accurate than a number of intuitive features.
Pulsed terahertz (T-ray) imaging systems represent an extremely promising method of obtaining sub-millimetre spectroscopic measurements for a wide range of applications. This paper investigates a number of techniques for optimally processing terahertz data. Specifically we consider wavelet de-noising and Wiener deconvolution algorithms.
A goal of this research is the design and implementation of a high speed, compact and portable T-ray imaging system. This system will draw heavily on MOEMS technology. A significant challenge in the development of such a system is the development of efficient software algorithms to perform signal recognition and imaging operations in real time. This paper takes the example application of a smart bio-sensor for surface tumours and investigates the signal processing techniques amenable to the tasks of efficiently de-convolving the system response, de-noising and extracting the salient features from the terahertz response waveform.