Undergraduates Mentored by McMillan and Weichert Present in Research Symposium

Posted on May 2023

The 25th Annual Undergraduate Research Symposium was held at Union South on Friday, April 28, 2023. The symposium offers an opportunity for undergraduate researchers from all areas of study to showcase their work in a professional setting, which helps students build public speaking and presentation skills. 

Research mentorship is a priority to the faculty in the Department of Radiology. Each year, our faculty mentor students and trainees of all levels – including undergraduates in this event. We are proud to highlight the work of these promising researchers, who were mentored by Alan McMillan, PhD and Jamey Weichert, PhD (as well as Zachary Miller, an SMPH MD/PhD student who completed his PhD with Kevin Johnson, PhD, and recently matched into Diagnostic Radiology for residency). 

Learn more about the projects presented at the Symposium below: 


Whole Body Segmentation of Computed Tomography Images 

Divya Durgavarjhula, mentored by Alan McMillan 

Identification of different organs via segmentation of medical imaging datasets is a necessary step in the development of automated analysis methods that augment physician’s capability to more quickly make accurate diagnosis and provide insights for personalized medicine. Artificial intelligence and deep learning approaches, such as convolutional neural networks (CNNs), have yielded promising results in segmenting computed tomography (CT) images with high accuracy, efficiency, and most importantly high speed. In this work, we describe a CNN approach for automated identification and segmentation of over 100 different body regions from whole body CT datasets via a deep learning method. 


Application of Radiomics to Disease Detection in Chest Radiographs 

Tracy He, mentored by Alan McMillan 

COVID-19 is still a trending topic in the biomedical imaging field. We investigated radiomics, a tool to study the texture features of images, using a machine learning approach to train classifiers which are able to output whether the patients have COVID from chest radiographs. The application of machine learning to radiology and medical imaging can be used as a digital co-pilot to help radiologists perform their job more effectively. Compared to other artificial intelligence methods, radiomics allows a potentially more interpretable approach. We investigated random forest models to demonstrate the performance of chest radiograph COVID classifiers using a varying number of features. 


Motion Correction MRI in the Lung 

Olivia Holsinger, mentored by Zachary Miller 

Image registration of time series with contrast dynamics have been explored in distinguishing between benign and malignant diseases. However, these images are often of poor quality as they are under-sampled. Several works have shown that integrating motion correction into image reconstruction significantly improves image quality, yet it is challenging to use traditional image registration algorithms on data with contrast dynamics. This is because typical loss functions are unable to distinguish between contrast change and motion, leading to poor image alignment. Through the use of a fully connected neural network and an external low dimensional signal, this study aims to explore motion correction of time series with contrast dynamics, where all motion field estimates come from signals completely independent of image contrast. 


Deep Learning Denoising for Magnetic Resonance Images 

Mahathi Karthikeyan, mentored by Alan McMillan 

The goal of this research is to improve the quality of medical images using deep learning methods. This is achieved by an algorithm designed to compare the original image with its degraded version (by injecting synthesized noise to the original image) and minimize the loss in quality between the two images. Specifically, MR images will be synthesized to generate three types of noisy images (salt & pepper noise, Gaussian noise, and Rician noise). A convolutional neural network will be used to develop improved quality images. 


Immune Checkpoint Inhibitors in Combination with Radiopharmaceutical Therapy 

Mason Kobxeeb Thao and Dane Paul Swenson, mentored by Jamey Weichert and Cynthia Choi 

Unlike many other diseases, cancer is unique with various categories of cancerous conditions requiring different approaches in therapy. Emerging approaches such as radiopharmaceutical therapy (RPT) and immune checkpoint inhibitors (ICI) have gained great recognition in recent years for their ability to selectively target cells accurately and effectively for treatment. Previous studies have shown a greater regression of cancer cells when radiopharmaceutical therapy and immune checkpoint inhibitors are administered jointly. The study investigates the aptitude of combined immune checkpoint inhibitors with radiopharmaceutical therapy using the monoclonal antibody Dinutuximab. Mice models are used to quantify the tumor volume of RPT compared to external beam radiation, ICI compared to non-ICI, and the control group to determine the therapeutic suitability of RPT in combination with ICI. 


Automatic Body Region Localization in Whole Body MRI Datasets Using Deep Learning 

Kevin Yuan, mentored by Alan McMillan 

This study aims to develop tools that can automatically localize body regions from whole body MRI datasets. This has the potential to introduce viable ways to utilize deep learning models in clinical workflows, specifically to help direct specialized algorithms to the proper locations in medical image datasets. A body region localization approach was developed for whole-body MRIs using axial slices as input. This study explored the use of various convolutional neural network structures to determine the best performance for an MRI body region localization task.