← Back to Event List

DE Seminar: Undergraduate Student Presentations

Location

Mathematics/Psychology : 401

Date & Time

May 2, 2022, 11:00 am12:00 pm

Description

Speaker 1: Michelle Ramsahoye (Advisor: Matthias Gobbert)

Title: 
Deep Fully Connected Residual Neural Network for Initial Energy Estimations of
Compton Camera Based Prompt Gamma Imaging Data for Proton Radiotherapy

Abstract: Proton radiotherapy uses a beam of protons to irradiate cancer tissue. It is an effective means of therapy whose dosage is characterized by the “Bragg peak”, which is a point of
maximum dosage with little to no remnant radiation beyond that point. One of the main
disadvantages is the possibility of radiation damage of healthy tissue surrounding the
cancer tissue. It has been suggested that real-time imaging can be used to help optimize
treatment delivery via prompt gamma ray image data collected with a Compton Camera.
When a prompt gamma deposits energy twice it is called a “double'' and physicists
assume that it has deposited all of its energy after the second collision thereby being
absorbed and the total of the energy depositions is considered to be the initial energy of
the prompt gamma ray. This initial energy is used during reconstruction to help determine
the origin of the prompt gamma and lead to a well formed reconstruction but the
assumption that a double deposits all of its energy is not always true leading to improper
origins based on incorrect initial energies. Here we present a deep residual fully
connected regression neural network which can make estimations of the initial energy of
double events on data generated by a Monte Carlo simulation using Compton camera
detector effects. We conduct a hyperparameter search to explore different models.
Performance was based on the smallest difference between training and validation loss
values, as well as overall lowest loss values. We then present the results of the currently
best performing regression neural network model, which demonstrate that this model
performs with a mean squared error training loss value of 0.0091 and validation loss
value of 0.0089. We suggest that further improvements such as increased model
complexity and capped energy values can help further reduce loss to an ideal value of
0.001 (or less) and improve model estimations for reconstructed images. This work was
done in collaboration with the Maryland Proton Treatment Center.



Speaker 2: Christina Dee (Advisor: Brad Peercy)

Title: CLUSTERED CELL MIGRATION: MODELING BOUNDARY POINT FORCES

Abstract: Cell migration is the progression of uni or multicellular units in response to chemical gradients or signals. It is an integral part of daily cellular functions; moreover, development and pathologies in migration can lead to discoveries in cancer metastasis. This research focuses on movement of a cluster of migratory border cells through nurse cells in the egg chambers of drosophila melanogaster, also known as the fruit fly. Due to its short life span, this organism is often used in research.The lab aims to answer: How does one represent membrane tension interactions in a complex mathematical model for individual cells? How can the lab combine these membrane forces with forces of migration and stochasticity to affect cell migration in a mathematical model for clustered border cells with complex boundaries?  Furthermore, what parameters are needed to correctly simulate cell migration?

The research utilizes MATLAB to simulate the egg chamber, border cells, and migratory cells. We captured heterogeneous cells of the egg chamber: including nurse, epithelial, and migratory border cells using interactions between cell membranes and arising via forces, including adhesive, repulsive, and spring forces. We developed the concept of neighborhoods to map interactions between cell boundaries. Furthermore, we used a volume force to include heterogeneously and realistically sized cells. After an initiation period where  cells filled the chamber, we attempt to enact the migratory force, which stems from a chemical gradient signaling, allowing border cells to climb through nurse cells. We solved the force balance equation with Euler's step to capture progression in time. In MATLAB, we coded cell boundaries around cell centers to represent membranes while adapting parameters, equations, and coefficients in order to realistically simulate cell migration. 

 While research is ongoing, the research team has been able to progress in an understanding of the model and further incorporation of more complexities, striving to achieve a simulation that accurately mirrors real life observation of the migratory cells.