Modeling & Imaging

OCT Simulator

A full frequency-domain OCT pipeline built from scratch — a Michelson-interferometer model, a simulated resolution phantom, spectral interferograms, and FFT depth reconstruction, with an analysis linking source bandwidth to axial resolution.

Axial resolution follows Δz = (2 ln2 / π)·λ₀²/Δλ — wider bandwidth, sharper depth.

source bandwidth Δλ · 100 nm

axial resolution 7.5 µm FWHM

two reflectors 12 µm apart — resolved

ToolsPython · FD-OCT · Fourier analysis

Read the report ↗

Brain Tumor Segmentation

“Tuning into Tumors” — a 2D-Fourier segmentation pipeline that treats each MRI slice as a signal, using high-pass and band-pass FFT filters (plus Canny edges) to isolate glioma boundaries, benchmarked against Otsu intensity thresholding.

Evaluated across 50+ annotated scans (UCSF preoperative-glioma set) on Dice, IoU, and boundary accuracy.
Frequency-domain filters recovered tumor edges that intensity thresholding missed.
Three-person student team; I built the FFT filtering and evaluation pipeline.

MRI slice: original beside FFT high-pass and band-pass reconstructions, with detected tumor boundary in red and the predicted centre marked by a green cross.

A second MRI slice showing FFT high-pass and band-pass tumor detection overlays with a predicted-centre green cross.

ToolsPython · 2D FFT · MRI · Dice / IoU · Canny

View poster ↗

CAR-T Tumor–Immune Modeling

“Cells Born to Kill” — an ODE model of CD19-directed CAR-T therapy in pediatric B-ALL, tracking malignant and normal B-cells, effector and memory CAR-T cells, and a lumped cytokine (CRS) compartment to link efficacy against toxicity.

Logistic tumor growth + CAR-T-mediated killing; memory cells reactivate with residual tumor burden.
Parameter sweeps over cytotoxic potency, memory reactivation, and effector proliferation revealed remission-vs-toxicity thresholds.
Framed CAR-T dynamics as a PK/PD-like system to relate dose and timing to outcome.

ToolsODE modeling · Python · PK/PD · Immunology

Read the report ↗