LEAD Lab neuroinformatics
Built ASR/NLP and fNIRS/fMRI pipelines that turn day-long home recordings and preschool neuroimaging into BIDS-compliant datasets for the SLIDE and PLANES studies.
The problem
The SLIDE and PLANES studies at the Language, Experience, and Development Lab collect preschool neuroimaging and day-long home audio. Raw fNIRS and fMRI are full of motion artifacts, the audio is hours of overlapping speech, and classroom streams do not arrive in a form that analysis code can trust. The scientific question is how early language experience shows up in the developing brain. The engineering problem is getting the data into a shape where that question can be asked.
What I did
I process preschool fNIRS and fMRI in Python and MATLAB to remove motion artifacts and align signals to standard brain anatomy. In parallel I build ASR and NLP pipelines for day-long home recordings, including speaker diarization and conversational turns, then fuse those multimodal classroom streams into BIDS-compliant datasets. I also automate RA quality-check workflows with documented Python scripts so the next person in the lab is not starting from a private notebook.
How it works
Imaging work is motion correction and anatomical alignment in Python and MATLAB. Audio work is speaker diarization and turn-taking on long home recordings. Fusion writes BIDS-compliant datasets so later analysis does not depend on ad hoc folder layouts. Quality-check scripts are documented and shared with graduate mentors and the PI (Dr. Rachel Romeo) as part of pipeline design, not as an afterthought.
What happened
The lab now has pipelines that turn messy preschool recordings and scans into analysis-ready BIDS datasets, with quality checks that RAs can run without reinventing the steps. That is infrastructure for the studies, not a one-off analysis of my own.