Preparation and Fast Quantum Control of 87Rb Bose-Einstein Condensates – MS (2020), Miami University, Ohio, USA.
This work consists of two main sections. In the first section, we present the main steps that were taken in the process of constructing a Bose-Einstein condensation (BEC) apparatus. A detailed description of the laser-optical system and vacuum system is provided. We also present the assembly and characterization of the first three-dimensional (3D) magneto-optical trap (MOT) in this double-MOT system. In the second part, we present numerical simulations on manipulating BECs at a fast rate while maintaining the coherence properties of its initial quantum state. Two-dimensional (2D) simulations of BEC transport are performed by numerically solving the Gross-Pitaevskii equation (GPE). In our simulations, we use trapping potentials in the form of painted potentials because it is possible to achieve arbitrary, dynamic traps with this method. To achieve high quantum fidelity, we use shortcuts-to adiabaticity (STA) for high-speed BEC transport. With these simulations, we compared different time intervals for a particular spatial displacement that a BEC can travel while keeping high quantum fidelity using experimentally feasible parameters.
Advisor: Dr. Carlo E. Samson
COMPUTERIZED MASS DETECTION SYSTEM FOR EARLY DETECTION OF BREAST CANCER – MSc in Medical Physics(2018), PGIS, University of Peradeniya, Sri Lanka.
The main objective of this research is to determine the most feasible artificial neural
network learning algorithm to get the highest prediction rate for mammographic mass
detection and to develop a low cost automated mammographic mass detection system for
early detection of breast cancer.
Detecting the masses of a mammogram can be a critical task due to the complexity of the
breast tissues. Radiographically masses can be described using their shape, margin, and
density. To extract the suspicious areas which can include masses from the mammogram,
adaptive threshold with the entropy and mean were used. These extracted areas are called
blobs. Five statistical features which interpret shape and texture features were calculated
from each of the extracted blobs. Then neural network training algorithm resilient propagation was used to model different neural networks since it has the highest prediction
accuracy, lowest error, and training time. After selecting the best ANN model for detecting
masses in mammograms using the mini-MIAS database the system was validated. To evaluate the mass detection system 121 extracted blobs were used in the neural network which includes 24 true masses. To train the system 96 blobs were used which included 18 true masses and to validate the system 25 blobs were used which included 6 true masses. All the true masses were detected and no false positives were marked when testing the system. The average training time was 359.26 ms and it was very low compared to other learning algorithms. Finally, 830 regions of interest from 86 mammograms were tested using the proposed system and 6 false positives were detected. The false-positive rate per image was 0.07. Compared to the previous studies the proposed method has high accuracy and
sensitivity.
Advisors: Profs. Roshan D. Yapa and Badra Hewavithana