| dc.description.abstract | Breast cancer has already invaded the women around the world with its brutal attack. Numerous patients are sacrificing their lives everyday due to lack of efficient
cure technology and a huge number of patients are still existing to hear the death
sentence. X-ray mammography is currently recognized as the golden standard of breast
cancer screening, but it suffers from high miss detection ratio, painful breast
compression and other harmful side effects. The main motto of this work is to identify
the tumor in its smallest dimension using an efficient, user-friendly and non-invasive
detection method without any side effects on human health. Towards this goal, a pyramidal shaped microstrip patch ultra wideband (UWB) antenna is proposed for frequency range of 3.23 GHz to 12 GHz for radar based microwave imaging system.
The performance of the antenna is measured in air media as well as in the vicinity of
breast model for lower band (3 GHz to 6 GHz) of UWB. In both cases, the antenna's
reflection (S11) and transmission (S21) coefficients are investigated in near field and far field region. A realistic breast model is also designed through Computer Simulation Technology (CST) software and experimentally. The distance between the transmitting antenna and breast model is varied from 1 mm up to 36 mm. The results show that, the proposed antenna performs better in near reactive region at a distance of 1 mm to 10 mm. Maximum and minimum transmission losses are -63.74 dB and -9.5 dB at 10 mm and 1 mm distance respectively. On the other hand, maximum and minimum reflection losses are found -1 dB and -52.58 dB at 36 mm and 2 mm respectively. In the whole experiment, the receiver is kept fixed at 1 mm apart from the breast and the received signals are reserved for the further signal processing. An efficient feature extraction technique (i.e., maximum, minimum, mean and standard deviation amplitude values of received pulse) is proposed here which also enhances the neural network training and testing performances by reducing the required time duration three times than previous studies. The overall system performance is verified by using proposed feature extraction and proposed antenna for various tumor sizes. The comparative study among support vector machine (SVM) kernel functions including linear function, radial basis
function, polynomial and multi layer perceptions are investigated and verified for
pattern recognition performance with 100% accuracy. SVM detects tumor size with
lesser accuracy than artificial neural network (ANN) results. The overall experimental system with ANN is able to detect tumor existence and tumor size of 1 mm (diameter) with nearly 99 % accuracy, which is around 3.2% more than related existing systems. | en_US |
