Research Experience

Systems Imaging & Bioinformatics Lab @ UM

Background
Characterizing the tumor immune microenvironment is essential for understanding the mechanisms of cancer progression and immune response. The development of spatially resolved technologies has revolutionized our ability to study the TME, necessitating the development of methods that can integrate spatial and cellular data to provide a comprehensive understanding of cell dynamics and drivers of response.

My work
I am developing multi-modal and spatial informatics methods to quantify the dynamics of the TME and discover biomarkers of response to treatment regimens. I am leveraging data such as spatial transcriptomics, multiplex immunofluorescence, and single-cell data in conjunction with methods in deep learning and statistics.

Wirtz/Wu Lab @ JHU

Background
Assessing histopathology slides is critical for diagnosing and researching cancer, but is challenging and time consuming for expert pathologists. While integrating deep learning in the field of pathology is promising, there are many challenges in data variability, annotation availability, and generalizability that stand in the way of adoption to the clinic.

My work
I developed a workflow for instance segmentation of immune cells and tissue subtypes in histopathology slides, with only sparse labels available. I collaborated with pathologists from Johns Hopkins Hospital to analyze precursor lesions of ovarian cancer.

Fertig Lab & Stein-O’Brien Lab @ JHU

Background
Non-negative matrix factorization is ideal for uncovering hidden patterns in single-cell data, but its implementation and interpretation can be challenging. This highlights the need for frameworks that are user-friendly, integrable, and efficient for such analyses.

My work
I created PyCoGAPS, a Python implementation of CoGAPS which is a Bayesian NMF algorithm for gene set analysis. PyCoGAPS enahnces runtime for large datasets, on the scale of tens of hours compared to CoGAPS, and additional workflows I developed with Docker and GenePattern facilitate user-friendly interpretation and implementation of NMF for single-cell analyses.

Precision Care Medicine @ JHU

Background
Static ocular torsion assessment is an important clinical tool for identifying abnormalities in the vestibular-ocular-motor pathway, but current methods are time-intensive with steep learning curves.

My work
I worked with a team of students to curate a synthetic, realistic dataset for static ocular torsion, and developed a classification model for assessing torsion from fundus images. We worked in collaboration with Johns Hopkins Hospital neurologists and researchers.

Malone Center for Engineering in Healthcare @ JHU

Background
Computer-assisted robotic surgeries have the potential to improve critical and common procedures, such as cataract surgeries, requiring fine precision and accuracy. A major component to its implementation involves the application of computer vision in order to detect regions of interest.

My work
I created a user-friendly script for annotating pupil segmentations across video frames of cataract surgical procedures. Following dataset curation, I implemented traditional computer vision methods and explored deep learning for segmentation of regions of interest.