Rice University data science researchers have introduced BRAINNET, a new AI software for brain tumor diagnosis and treatment. Their paper, Glioblastoma Tumor Segmentation Using an Ensemble of Vision Transformers, introduces software that segments brain MRIs, providing a more efficient, highly-effective method for brain cancer detection. But notably, these researchers are not PhD students, postdocs, or professors — they are working professionals.
The paper, which was presented at the SPIE Medical Imaging 2025 conference in February, grew out of a capstone project for the professional masters of data science (MDS) program. The authors were Chevron Digital Scholars: employees on a one-year education leave of absence to complete the MDS degree.
Arko Barman, assistant teaching professor in Rice’s Data to Knowledge (D2K) Lab, supervised the capstone project along with faculty member Xinjie Lan. Upon review of the team’s work at the end of the semester, he found the scholarship rigorous enough to be a potential research article.
“Having a research paper from coursework is really rare,” said Barman, who presented the paper at SPIE Medical Imaging. “It's usually PhD students and postdoctoral researchers who complete research projects to write papers and publish them. Having a team of master’s students doing this research mostly in the timeframe of just one semester is really saying something about how much they learned and how much hard work they put in.”
Choosing brain MRIs as the focus
Barman said that the Scholars wanted to tackle an important, healthcare-related topic for their MDS capstone project. Key contributor and project lead Huafeng Liu, a senior technical geophysicist at Chevron, sought a capstone project that would use the skills he acquired as a Digital Scholar and would benefit his work as a geophysicist. The team, which also included Rice MDS co-authors Benjamin Dowdell, Todd Engelder, Zarah Pulmano, and Nicolas Osa, aimed to find a topic that could also contribute positively to society.
The group decided to focus on medical imaging — specifically, 3D brain MRIs — since they had easily accessible raw data. They used the UPennGBM dataset for their research, consisting of 3D multi-parametric MRI (mpMRI) scans from 611 subjects. Another reason they chose the subject was the similarity between interpreting MRIs and interpreting seismic images.
“Inspired by the capabilities of computer vision models, I decided to explore their potential application to 3D seismic images, which I frequently work with in my job,” said Liu. “The abundance of 3D images in medical imaging and the similarities in identifying regions of interest between medical and seismic images made this an ideal task. Additionally, the potential to impact a broader community through my research was highly motivating. Experimenting with computer vision models on 3D MRI data seamlessly connected these dots.”
MDS and Digital Scholars
The MDS program — a professional, non-thesis master’s program geared toward interdisciplinary working professionals seeking data science skills — is administered by Rice’s Department of Computer Science. After completing the program, industry professionals have mastery in a core area of data science and the ability to turn raw data into actionable insights.
"The Rice MDS degree is designed to prepare future leaders in data science,” said Scott Rixner, Rice CS’ director of professional studies. “Our instructors take pride in their mission to prepare students for cutting-edge work in data science, not only by providing them with strong foundational knowledge but also by pushing them to innovate. With this encouragement, our students are able to go above and beyond in their capstone projects to discover new and exciting insights in important areas."
Through Chevron’s Digital Scholars program, select employees could enroll in the MDS program and apply their skills to accelerate the company’s integration of digital tools and to identify opportunities for innovation.
Liu said, “Chevron’s Digital Scholar program provided me with an exceptional opportunity to learn data science systematically from experts in the field and to practice my skills through hands-on examples.” Obtaining his MDS through the Digital Scholar program “increased the visibility of my skills,” he said, and led to “an opportunity to work on one of the most challenging and impactful problems in hydrocarbon exploration.”
D2K: Emphasizing experiential learning
The culmination of the MDS program is the capstone project, offered and administered by the D2K Lab. In the capstone, MDS students start with a raw, uncurated data set and use their new skills to clean, organize, and analyze the data to address a real-world data-science challenge.
Barman described the importance of this process. “I think there has been traditionally a disconnect between lecture-style courses and the theoretical concepts taught in these courses and what happens in the real world. One of the goals for the capstone course and the D2K lab is to bridge this gap.”
The D2K Lab is not like a conventional research lab,” he explained. “The D2K Lab is a teaching-focused center that tries to provide students with experiential learning. They learn through actually doing stuff. It's not like lecture-style learning; it’s working on a project and learning through experience.
We do not work with clean, curated datasets. We work with raw, dirty datasets from the real world.”
Liu benefited from D2K’s hands-on approach in several ways.” He said, “The student-centric approach of the capstone project was instrumental in fostering innovation. Team members could share various perspectives. This collaborative environment encouraged the exploration of multiple possibilities through creative thinking.”
He said D2K instructors, like Barman, “played a crucial role.” They asked important questions, offered ideas to address challenges, provided feedback on feasibility, and fostered teamwork.
Significance of the research
“Glioblastoma is a particularly aggressive form of brain cancer; it's almost like a death sentence,” said Barman, who has a background in computer vision and medical image analysis. He explained that clinicians rely on MRIs for glioblastoma treatment planning and diagnosis and need to segment the tumor region in the MRI and get “the exact location and boundary” of the tumor. Radiologists spend six to eight hours a day, 255 days a year, working on tumor segmentation, “which potentially causes fatigue,” he said.
“Once our brains get fatigued, we are prone to sacrificing efficacy and efficiency, thus making more medical errors. What we tried to do is develop an AI algorithm to do these segmentations used for treatment planning,” said Barman. This resulted in the Brain Radiology Aided by Intelligent Neural NETworks (BRAINNET) tumor segmentation model.
“The BRAINNET system is designed to assist radiologists in quickly identifying the presence and severity of glioblastoma,” said Liu. “By enhancing the accuracy, efficiency, and effectiveness of brain tumor diagnosis and treatment, BRAINNET aims to significantly improve patient care and outcomes.”
Barman agreed that BRAINNET was a step forward in image segmentation. “What they actually accomplished, the models they applied, are really cutting edge,” he observed. “And the performance of these models is better than other state-of-the-art models in the literature.”
Liu said he and his project team “hope the outcome of this research can be integrated into tools used by radiologists to assist in diagnosing and treating glioblastoma patients. Additionally, we aim for this research to inspire other researchers in the medical imaging AI domain to further advance their work. By improving diagnostic tools and encouraging innovation, we can contribute to better patient outcomes and advancements in medical technology.”
The BRAINNET software is available for public use via Rice D2K’s GitHub site.
For more information about Rice University’s applied Master of Data Science program, offered both 100% online or on-campus, please visit Rice’s MDS website.
More information about Rice’s Data to Knowledge Lab can be found on the D2K website.