Dielectric modelling of human skin and breast tissue in terahertz frequencies : potential application to cancer detection
- Publication Type:
- Thesis
- Issue Date:
- 2015
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Growing developments in the generation and detection of terahertz (THz) radiation over more than two decades have created a strong incentive for researchers to study the biomedical applications of terahertz imaging. Contrasts in the THz images of various types of cancer, especially skin and breast cancer, are associated with changes in the dielectric properties of cancerous tissues. In fact, dielectric models can explain the interaction between terahertz radiation and human tissue at a molecular level just as their parameters have the potential for becoming indicators of cancer. However, dielectric modelling of various forms of human tissue remains limited due to a number of factors, especially suboptimal fitting algorithms and tissue heterogeneity.
Thanks to the high water content of human skin, its dielectric response to terahertz radiation can be described by the double Debye model. The existing fitting method using a nonlinear least square algorithm can extract the model parameters which track their measurements accurately at frequencies higher than one THz but poorly at lower frequencies. However, the majority of dielectric contrast between normal and cancerous skin tissues has been observed in the low THz range. Accordingly, this research has developed two global optimization algorithms which are capable of globally accurate tracking thereby supporting the full validity of the double Debye model in simulating the dielectric spectra of human skin in the THz frequencies. Numerical results confirm their superiority over the conventional methods. Furthermore, the next goal of the study is to apply statistical analysis to the parameters of the double Debye model in order to test their discrimination capability of skin cancer from normal tissue. Linear programming and support vector machine algorithms have also been employed using these parameters to classify normal skin tissue and basal cell carcinoma. By combining the double Debye parameters, the classification accuracy has shown significant improvement. The encouraging outcomes confirm the classification potential of the double Debye parameters.
The double Debye model, however, has been shown to be not suitable for simulating human breast tissue due to its low water content and heterogeneous structure, thus limiting the understanding of the THz dielectric response of breast tissue. To overcome this problem, this study proposes a new non-Debye dielectric model to fit the dielectric spectra of human breast tissue. Due to the mathematical complexity of the fitting procedure, a sampling gradient algorithm of non-smooth optimization is used to optimize the fitting solution. Simulation results confirm applicability of the non-Debye model through its exceptional ability to fit the examined data. Statistical measures have also been used to analyse the possibility of using the parameters of this model to differentiate breast tumours from healthy breast tissue. Based on the statistical analysis, popular classification methods such as support vector machines and Bayesian neural network have also been applied to examine these parameters and their combinations for breast cancer classification. The obtained classification accuracies indicate the classification potential of the model parameters as well as highlighting several valuable features of the parameter combinations.
Please use this identifier to cite or link to this item: