ZAHRA FARSHIDROKH

Analyzed multivariate time-series features in perturbations for short- and medium-horizon forecasting, focusing on structural dependencies over black-box accuracy. Studied system dynamics, external shocks, regime changes, and cross-system interactions. Designed and implemented a model interpretability and diagnostics framework using SHAP, LIME, and feature attribution methods. Performed feature ablation and baseline experiments to quantify contributions across regimes. Evaluated forecasting under heterogeneous conditions and communicated findings through technical reports and stakeholder presentations.

Processed ECG/SCG signals using wavelets and filtering to support clinical imaging analyses. Developed real-time motion tracking algorithms for integration into MRI workflows. Collaborated with clinicians and physicists to refine methods for clinical applicability. Translated electrophysiological signals into quantitative metrics applicable to in-vivo neural research.

Developed, trained, and evaluated deep learning models (U-Net, VAE, CNNs) for medical image segmentation, registration, reconstruction, and classification. Built reproducible imaging pipelines integrating classical algorithms (Elastix, SimpleITK) with modern AI frameworks (PyTorch, MONAI). Performed quantitative validation using Dice, Hausdorff, MI, and NCC, and optimized models for clinical-grade reliability. Collaborated with clinicians and imaging scientists to translate research models into deployable tools.