Evan Grove ('26), a System Science & Engineering major, spent this past summer serving as a Data Science Intern at the Children's Hospital of Philadelphia, working in the Translational Imaging Research Unit in the Center for Data Driven Discovery in Biomedicine. In this internship Evan worked to develop and apply artificial intelligence to better classify pediatric brain tumors. This experience was supported by the Franklin Opportunity Fund, an internship funding opportunity for Benjamin Franklin Scholars.
Evan Grove – Engineering ’26 | Systems Science & Engineering
This summer, I worked as a Data Science intern at the Translational Imaging Research Unit in the Center for Data Driven Discovery in Biomedicine (D³b) at the Children’s Hospital of Philadelphia. My project focused on developing and applying artificial intelligence methods to whole-slide pathology images in order to classify subtypes of medulloblastoma, the most common malignant pediatric brain tumor. This work was especially meaningful because medulloblastoma subtyping can inform prognosis and treatment strategies, making improved classification clinically relevant to patient outcomes.
My role began with data preprocessing and cleaning, which quickly taught me the importance of careful pipeline construction in computational pathology. I worked with raw whole-slide images, which are very large files. Using QuPath, I implemented a tiling script to divide each slide into smaller tiles and filter them to retain only tumor regions. I then applied a nuclei detection script to identify relevant cellular structures within each tile. Much of this pipeline was organized and managed within Flywheel, the lab’s data platform, which streamlined the storage of data and workflows across the research team.
Once preprocessing was complete, quantitative pathology features were extracted for each patient. Our lab also had an existing dataset of radiomic features derived from MRI scans. Combining the radiomic and pathomic data created a rich multimodal dataset, which I analyzed in Google Colab using a Support Vector Machine (SVM) classifier with leave-one-subject-out (LOSO) validation. I experimented with feature selection and observed that radiomic features performed best when reduced to the top 4–5%, whereas pathomic features improved steadily as more were included. When the optimized radiomic and pathomic features were merged, the combined model achieved an accuracy of 83.9% in distinguishing SHH/WNT medulloblastoma subtypes from Group 3/4 subtypes.
Pictured: Evan with the D3b Translational Imaging lab Team
One of the most valuable things I gained from this experience was a much deeper understanding of how preprocessing and feature selection directly influence downstream model performance. I became more comfortable handling large-scale biomedical data and gained practical exposure to tools like QuPath and Flywheel. I also learned how computational and biological expertise come together in a research environment to generate clinically meaningful findings. At the same time, not everything went smoothly. Through a long process of model tuning, I learned that performance often declined when I added too many features, which pushed me to think more carefully about data quality rather than just quantity. These challenges forced me to problem-solve, seek guidance, and adapt my workflow early on.
The project also had a clear impact on how I think about research. It showed me how methodological decisions that may seem technical or routine can shape the usefulness of a model in a clinically relevant setting. It also strengthened my ability to communicate results to collaborators with different backgrounds and taught me how to stay resilient when experiments did not initially work as expected.
Looking ahead, this internship reinforced my motivation to pursue biomedical data science as part of my academic and career goals. Seeing firsthand how artificial intelligence could support better classification of pediatric brain tumors strengthened my desire to use data science to address healthcare challenges and improve patient outcomes.
Interested in reading more accounts from BFS? Click Here
Related Articles
Connections and Community
Hannah Shumsky (‘26), a Neuroscience major, conducted research on the effect of circadian rhythms on lung repair and regeneration with mentorship from Dr. Shaon Sengupta (Division of Neonatology, Children’s Hospital of Philadelphia). This research was supported by the College Alumni Society Undergraduate Research Grant.
Blending Scientific Training with Advocacy and Business Strategy: Interning at Saper Law
Vivi Sankar ('27), a double major in Chemistry and Finance, spent this last summer working as a Law Clerk at Saper Law, a firm that focuses on intellectual property and business law. The internship reinforced Vivi's goal of becoming a patent lawyer, as she plans to use her scientific background and business acumen to strengthen her legal career. This experience was supported by the Franklin Opportunity Fund, an internship funding opportunity for Benjamin Franklin Scholars.