Differential Equations Seminar: Undergraduate Researchers
UMBC Undergraduate Math Students Present Virtually
Location
Mathematics/Psychology : 401
Date & Time
May 4, 2020, 11:00 am – 12:00 pm
Description
Speakers:
Jonathan Basalyga (Advised by Dr. Gobbert)
Title: Improving Compton Camera Data through Deep Learning for the
Reconstruction of Proton Beams in Cancer Treatment
Abstract: Real-time imaging has potential to greatly increase the effectiveness
of proton beam therapy for cancer treatment.
One promising method of real-time imaging is the use of a Compton
camera to detect prompt gamma rays, which are emitted by the beam,
in order to reconstruct their origin.
However, because of limitations in the Compton camera's ability to
detect prompt gammas, the data is often ambiguous,
making reconstructions based on it unusable for practical purposes.
Deep learning's ability to detect subtleties in data that traditional
models do not use make it one possible candidate for the improvement
of Compton camera data.
We show that a suitably designed neural network can
reduce false detections and misorderings of camera events,
thereby improving reconstruction quality.
This is joint work with Gerson C. Kroiz, Carlos A. Barajas, and
Matthias K. Gobbert in the department and Paul Maggi and Jerimy Polf
from the University of Maryland School of Medicine.
This work is partially supported by an REU Supplement to
the NSF-funded CyberTraining initiative at UMBC.
Reconstruction of Proton Beams in Cancer Treatment
Abstract: Real-time imaging has potential to greatly increase the effectiveness
of proton beam therapy for cancer treatment.
One promising method of real-time imaging is the use of a Compton
camera to detect prompt gamma rays, which are emitted by the beam,
in order to reconstruct their origin.
However, because of limitations in the Compton camera's ability to
detect prompt gammas, the data is often ambiguous,
making reconstructions based on it unusable for practical purposes.
Deep learning's ability to detect subtleties in data that traditional
models do not use make it one possible candidate for the improvement
of Compton camera data.
We show that a suitably designed neural network can
reduce false detections and misorderings of camera events,
thereby improving reconstruction quality.
This is joint work with Gerson C. Kroiz, Carlos A. Barajas, and
Matthias K. Gobbert in the department and Paul Maggi and Jerimy Polf
from the University of Maryland School of Medicine.
This work is partially supported by an REU Supplement to
the NSF-funded CyberTraining initiative at UMBC.
Gerson Kroiz (Advised by Dr. Gobbert)
Title: A Comparison Of Stochastic Precipitation Generation Models For The Potomac River Basin
Abstract: Weather ensembles are an integral part of weather forecasting, and can also be used to test the sensitivity and performance of climate models.
Among meteorological variables, simultaneous simulation of precipitation
at multiple sites presents unique challenges since precipitation has a
semi-continuous distribution. We compare Robertson's Hidden Markov Model
setup with Wilks' Multivariate Markov Chain based generator to see how
well they recreate the spatiotemporal characteristic of gridded satellite
precipitation estimates. Our results show that the Wilks method does a
better job of capturing spatial correlations while the HMM model can
estimate and simulate longer durations of time. This work was done in
collaboration with Jonathan N. Basalyga, Uchendu Uchendu, Carlos A.
Barajas, Reetam Majumder, Matthias K. Gobbert and Nagaraj K. Neerchal in
the department as well as Kal Markert from Alabama-Huntsville and Amita
Mehta from JCET. This work is partially supported by an REU Supplement to
the NSF-funded CyberTraining initiative at UMBC as well as an
Undergraduate Research Award (URA) from UMBC.
Ashley Copenhaver (Advised by Dr. Peercy)
Title: Understanding Connectivity of Pancreatic Beta Cells through Artificial Neural Networks
Abstract: The islet of Langerhans consists of hundreds of beta cells whose synchronization is key to the proper secretion of insulin from the endocrine component of the pancreas. Experiments have suggested the existence of a type of β-cell in the islet, called the hub cell, which controls islet synchronicity. If silenced, the hub cell appears to desynchronize the islet. Simulations based on the experimental data have not confirmed the proposed high functional connectivity of hub cells. Instead, we have used numerical exploration to show the existence of a similar β-cell, termed the switch cell, which can control the activity of the islet without needing functional connectivity. We are using artificial neural network techniques to identify islets containing switch cells based upon cell characteristics and cell-coupling values. We begin with a two-cell network using three parameters to identify a switch islet. When we are able to identify switch islets, we begin working with a three-cell network in six parameters and continue to scale up to a 57-cell network. We aim to discover what biophysical features are important for the algorithm to produce good results and whether the algorithm’s boundary line converges to the theoretical boundary line predicted by simulation results.
Tags: