Skip to content

Shepherd Research Lab Blog

Summer interns for the 2022 CREATE Program present their projects

Yimin discussing her poster with a faculty member.

The CREATE Program at the University of Hawaiʻi Cancer Center concluded today with poster presentations for all participants. These research projects took place from June to July.

Shepherd Research Lab was happy to welcome two interns this year: Christian Leidholm and Yimin Yuan.

Christian, a junior in Data Science at Purdue University, explored the Federated Learning method to create an AI model for classifying breast ultrasounds. This method allows clinical partner sites to contribute to ultrasound cancer classification model building without sharing sensitive patient information. The resulting models could then be used with portable ultrasounds to increase access to breast cancer screening in high-risk areas such as the US-Affiliated Pacific Islands.

Yimin is currently a student at Kapiolani Community College studying General Engineering with the plan to transfer to the University of Hawaiʻi’s Biological Engineering program. Her research focused on identifying unique characteristics of obesity in children using DXA scans from our Shape Up! Kids study.

In just two months, these projects added significant contributions to ongoing SRL studies. Christian says, “It feels great to contribute to further cancer research, what we’re doing really helps.”

SRL 2022 CREATE Projects

Christian Leidholm, Thomas K. Wolfgruber, Lambert Leong, Arianna Bunnell, John A. Shepherd. Federated Learning for malignant cancer classification by AI in breast ultrasound images while retaining patient privacy at clinical sites. Poster presented at: CREATE Summer Internship Poster Presentation; 2022 Jul 29.

This federated learning network was created using three identical computers—one federated learning server and 2 client servers imitating different sites. A data set of 210 malignant breast ultrasounds and 620 normal and benign images was split between the two clients. Each trains a local AI model that is then sent and averaged by the federated learning server. This is what allows the federated learning method of AI training to keep sensitive patient information private at each clinical site.

Christian documented the process of setting up a federated learning server, which he built in collaboration with other SRL members.

“It was great to work with such skilled coworkers,” he says. “What I worked on, I’ve never worked on before but I learned a lot. It was something very cutting edge.”

Federated learning was then compared to the traditional centrally trained AI model for efficiency. The tests show that the federated learning model was slightly less effective at detecting malignancies. However, the model was created faster and the accuracy improved with each round of training. This shows great potential. With further testing, the federated learning model may be an option for quickly creating AI-assisted ultrasounds for regions with less access to breast cancer screening.

Yimin Yuan, Michael C. Wong, Aimee Bowen, En Y. Liu, Nisa N. Kelly, John A. Shepherd. Defining Obesity Characteristics in Children Using DXA Shape and Appearance Modeling. Poster presented at: CREATE Summer Internship Poster Presentation; 2022 Jul 29.

Using DXA scans from the Shape Up! Kids study, this project sought to define unique phenotypes of children in order to better describe obesity beyond conventional BMI. This was done using DXA shape and appearance modeling, a method also used with our Shape Up! Adults data.

The data included 396 healthy children (222 female and 174 male) between the ages of 5 and 18 years. Yimin performed principal component analysis on the data before using K-means clustering to classify each participant. Using logistic regression, the unique cluster characteristics were identified. These results are a great stepping stone for further research into childhood obesity characteristics.

“I learned so many things,” Yimin commented. “The most important thing I’ve learned is about Biomedical Imaging, one of the concentrations of my intended major at UH. Also, I’ve learned about coding. It was amazing!”

The CREATE Program at UHCC

UH Cancer Center annually offers undergraduate students the opportunity for hands-on summer research experiences. Participants were paired with a mentor from the Population Sciences in the Pacific department or the Cancer Biology department. Along with individual mentorships, there were multiple activities for all interns which allowed them to meet others in the program.

“I had the opportunity of meeting people with the same interests as me: cancer research,” said Yimin.

Christian agreed, saying, “You really do learn a lot [and it was] great to meet other students my age with similar interests.”

2022 CREATE summer poster session.

These projects had a real impact on the work at UH Cancer Center and SRL is thankful for the hard work of these students.

Share with your friends