Intelligent Microwaves-Based Modalities for Breast Cancer Detection
MetadataShow full item record
Breast cancer is considered to be one of the major causes of mortality in women worldwide. Detection of breast tumors in their early stage is the key factor for possible successful treatment and can significantly reduce mortality rates. In recent years, microwaves have emerged as a potential technique for breast cancer detection one that avoids the discomfort, risks and costs associated with x-rays and excessive cost and availability of MRI. Microwave technique is simpler to use, much less expensive to generate, and is non-ionizing. The microwave detection used in earlier works relied on the sharp contrast in the electrical properties between tumors and healthy tissue. In such methods, the breast was scanned by microwaves of various frequencies and the reflection recorded. An image depicting the electrical properties of the breast was then developed. The challenge, however, is that female breasts contain a complex network of fat and fibrous tissues, the electrical properties of which can very well resemble those of cancerous or benign tumors. Also, the electrical properties of the breast vary with frequency, requiring the earlier techniques to employ complex receptors. Motivated by these drawbacks, this thesis addresses the development of an inexpensive, non-ionizing and highly sensitive microwave technique for detecting early-stage breast tumors. In the first part of this dissertation, anatomically-realistic numerical breast phantom models are constructed using computer simulation technology (CST). The phantoms are anatomically realistic three dimensional (3D) numerical models that are realistic in both structural and dielectric properties. In the second part of the thesis, first a single electric probe and then a magnetic probe are individually combined with classification algorithms to help in detecting the presence of breast tumors. A key feature of our proposed detection concept is the almost simultaneous sensing of both a woman breasts, since right and left healthy breasts are morphologically and materially identical except amongst very small percentage of women. The two tests then can be compared to reveal any tissues property discrepancies. The concept employs a near-field resonant probe with an ultra-narrow frequency response. The resonant probe is highly sensitive to any changes in the electromagnetic properties of breast tissues, such that the presence of a tumor can be gauged by determining the changes in the magnitude and phase response of the sensor's reflection coefficient. Once the probe response is recorded for both breasts, Principle Component Analysis (PCA) method is employed to emphasize any difference in probe responses. For validation of the concept, tumors embedded in realistic breast phantoms were simulated. To provide additional confidence in the detection modality introduced here, experimental results of three different sizes of metallic spheres, mimicking tumors, inserted inside chicken and beef meat were detected, first by using an electric probe and then using a magnetic probe, operating at 200 and 528 MHz respectively. The results obtained from the numerical tests and experiments strongly suggest that the detection modality presented here might lead to inexpensive and portable modality for early and regular breast tumor detection. A novel modality proposed in the third part of the thesis significantly enlarges the sensitivity area beyond that of a single probe. This modality, based on a sensor we developed, relies on a 4-element identical antenna array fed with a single port. The use of this senor array improves the sensitivity area as compared to a single sensor, resulting in better detection of tumors located deeply inside breast tissues. Two different sensors are developed in this part,a dipole sensor and a loop sensor. The dipole sensor comprises a 4-element identical dipole antenna array fed with a single port. Numerical simulations have been conducted using a numerical breast model with and without tumor cells placed in the near-field of the sensor. The sensor is capable of detecting a breast tumor inserted at four different locations and of various sizes. Experimental validation was conducted using chicken meat and metallic and dielectric spheres that resemble healthy and tumourous breast tissues. The simulation and experimental results show that the array sensor has a high sensitivity for detecting various sizes of breast tumor inserted at different locations. The developed loop sensor comprises a 4-element identical loop antenna array fed with a single port. Numerical simulations have been conducted using a numerical breast model with and without tumor cells placed in the near-field of the sensor. The sensor is capable of detecting various sizes of tumor inserted at five different locations. Experimental validation was conducted using a glass box filled with vegetable oil and metallic spheres that resemble healthy and tumourous breast tissues, respectively. The simulation and experimental results show that the array sensor has a high sensitivity for detecting a metallic sphere placed at five different locations inside a dielectric medium as well as for detecting variable sizes of metallic sphere. In the fourth part of this thesis, a near-field metasurface sensor is introduced whereby a near-field array sensor operating in the microwave regime is used statically to identify the presence of a breast tumor. In a departure from conventional near-field sensors, the sensor is a metasurface comprising an array of 8$\times$8 electrically-small resonating elements. The elements of the metasurface are designed to respond to both electric and magnetic fields. This capability enables the metasurface to emphasize seemingly small changes in the composition of the electric and magnetic fields in its environment, thus leading to higher overall sensor sensitivity. Furthermore, unlike previous near-field probes, the overall metasurface sensor is not electrically small, which means that it provides a larger sensing surface while maintaining the effectiveness of near-field probes in the sense of detecting material changes in the near proximity of the sensor. Numerical and experimental tests were used to validate the proposed detection methodology. This was achieved by testing the metasurface with a breast phantom having tumor placed at single location at three different stand off distances and with a breast phantom having tumors placed at different locations. Measurements were carried out on a realistic phantom that mimic a real female breast in terms of electric properties. The results showed high sensitivity of the metasurface which can indicate the existence of an anomaly that resembles a tumor inside a breast phantom having inhomogeneous material composition. The advantage of the proposed metasurface sensor array as compared to previously introduced sensors is that the proposed array sensor is fed by a single-feed point. Unlike multiple-feed points sensors, this single feeding port sensor array significantly reduces the computational cost and complexity caused by processing the data from multiple feeds. The thesis then discusses the idea of using machine learning approaches to improve the performance of the proposed microwave detection system. The machine learning methods proposed discriminated between normal and abnormal breast phantoms in different sizes and classes of breasts, then also significantly improved the accuracy, sensitivity and specificity of the proposed detection system. As future work, the last part introduces several ideas for solving challenges in various aspects of the proposed sensors and the classification logarithms introduced in the developed system. The first idea is introduced to improve the sensitivity of the metasurface sensor by using multiple polarization sensors. The metasurface sensor, presented in chapter seven has one diploe in the middle of the loop, which will be extended to have two cross dipoles for vertical and horizontal polarization excitations. The second idea is to improve the sensitivity area of the proposed system by using multiple metasurface sensors that cover the whole breast and therefore eliminate the use of mechanical motors to move the sensor all over a breast. The third idea is to develop a portable detection system and integrate of the standalone VNA and the sensor into one miniaturized unit. The VNA circuitry will be positioned at the back of the sensor and will be connected with a laptop.
Cite this version of the work
Maged Aldhaeebi (2020). Intelligent Microwaves-Based Modalities for Breast Cancer Detection. UWSpace. http://hdl.handle.net/10012/15696