Breast Cancer Detection Using Microwaves and Microwave-Thermography Techniques
Loading...
Date
2021-11-16
Authors
Alsaedi, Dawood
Advisor
Ramahi, Omar
Journal Title
Journal ISSN
Volume Title
Publisher
University of Waterloo
Abstract
The accelerated growth in Microwave Imaging (MWI) and Microwave Detection (MWD)
is driven by microwave ability to penetrate materials that are considered opaque at a
shorter wavelength. This elevates MWI potential to cover a wide range of applications,
including but not limited to security checking; civil and industrial operations; and medical
diagnostics. Involving microwaves in the biomedical eld relies on two additional microwave
features: a) a microwave interacts differently with biological tissue, based on the tissue's
electrical properties; and b) a microwave is considered as non-ionizing radiation, thus it
presents low to no risk to biological tissue. This has instigated the implementation of microwaves
in different areas of biomedical diagnostics, such as brain haemorrhages, meniscus
tears, and breast cancer detection.
The first part of this dissertation presents a microwave thermography hybrid breast
cancer detection technique consists of a microwave radiation source, an infrared heat detector,
and a machine learning algorithm. Since many conventional MWI approaches collect
transmission signals along with re
reflected signals, this technique is based on recording the
electromagnetic wave after passing through the entire breast. A sensitive lm is placed behind
the breast (opposite direction of the radiation source) to absorb the transmitted wave
that propagates according to the dielectric properties of the breast tissue. The captured
heat pattern is used as a guide for determining the presence of an anomaly within the
breast tissue. Machine learning is used to enhance the detection accuracy and to provide
further information about the tumor's features, such as size and location. The proposed
modality shows a capability to detect and determine the size and location of an artificial
tumor with a 5 mm radius and a 2:1 permittivity contrast with normal tissue.
A new breast cancer detection modality that uses a metasurface as the imaging medium
and a microwave radiation source is introduced in the second part of this thesis. In contrast
with previous microwave imaging techniques, or imaging techniques in general, instead of
providing an image of the internal breast tissue (i.e., a slice/cut through the breast), the
proposed technique provides an impression of the breast tissue. The impression which, in
principle, is similar to the impression captured by an x-ray lm, is captured by a metasurface,
which is an ensemble of electrically-small resonators. Each cell records the strength
of the incident power that impinges on it. This metasurface may be viewed as analogous to
a low-frequency scaling of the x-ray lm. The metasurface receives the transmitted energy
through the breast, which resembles the mammography approach. While mammography
faces a major challenge in detecting tumors in dense breasts, the metasurface proposed
here utilizes a low-frequency radiation source, thus allowing higher penetration through
dense breast tissue. Similar to the first part of the thesis, by building a proper machine learning code, the detection capability is then enhanced by engaging a Convolution Neural
Network (CNN) to determine the tumor's features.
In the third part of this thesis,
flexible-conformal metasurface films are introduced for
utilization as a wearable breast cancer detection modality. This technique involves two
metasurface arrays with electrically small and closely spaced resonators as a transmitter
and receiver. The microwave radiation of the metasurface transmitter is received at the
other side of the breast by the metasurface receiver to form an image of the received
electromagnetic power. The receiver is expected to provide an electromagnetic energy
pattern instead of an image that represents the breast's internal contents. The power
pattern is influenced differently according to the electrical properties of the breast tissue,
thus a unique power impression can be obtained for both a healthy breast and a breast
that contains a cancerous tumor. By using
flexible metasurface films, the antenna number penetration-
resolution trade-off that limits the capability of conventional MWI techniques
can be minimized. In addition, the necessity for antenna miniaturization to enhance the
resolution can be avoided, and the long mechanical scan can be replaced with just a few
scanning steps of the metasurface sheet.
Description
Keywords
microwave imaging, breast cancer detection, convolution neural network, medical imaging, machine learning, metasurface