Tricia Lobo

Computational researcher developing patient-specific cardiovascular modeling pipelines from 4D medical imaging, with research spanning nnU-Net-based automated segmentation, deformable image registration, finite-element-based strain validation, Bayesian mesh optimization, and Laplace–Beltrami methods. Interested in extending these methods toward fluid–structure interaction, cardiac digital twins, and human-relevant computational models for translational research.

Staff Researcher - Scientific Computing & Cardiovascular Modelling - University of Pennsylvania Image Computing and Science Laboratory - Philadelphia, PA

(2025-01)

• Developed an end-to-end patient-specific cardiovascular modeling pipeline from ECG-gated 4D CT, integrating automated segmentation, deformable image registration, remeshing, finite-element validation, and regional strain analysis.
• Trained and evaluated a 3D nnU-Net for automated ascending-aorta segmentation from ECG-gated 4D CT; integrated model predictions into the downstream strain pipeline and quantified agreement with observer-derived strain maps.
• Designed and implemented an automated finite-element-based strain-validation framework to tune diffeomorphic registration parameters across patients, segmentation sources, and imaging systems; identified shared parameter regions supporting robust strain recovery.
• Developed Bayesian mesh-optimization workflows using Optuna to tune remeshing parameters against geometric quality metrics and improve downstream numerical stability.
• Applied Laplace–Beltrami spectral decomposition to characterize strain fields across spatial scales and quantify segmentation-sensitive components of regional disagreement.
• Evaluated registration and strain-recovery performance using spatial correspondence, Bland–Altman analysis, correlation, and multiscale error analysis across manual and deep-learning-derived segmentations. Achieved held-out nnU-Net Dice scores up to 0.99 and strong downstream strain agreement with observer-derived maps.
• Built an EpiNet-augmented transformer for uncertainty-aware prediction of finite-element aortic strain fields, preserving predictive accuracy while modeling spatially coherent epistemic uncertainty.

Software Engineer - Candid

(2022-06 - 2024-10)

• Owned end-to-end ML infrastructure for orthodontic image classification, from model training through cloud deployment.
• Built AWS-based infrastructure to accelerate model iteration and deployment.
• Developed Redis-backed distributed processing systems, increasing computational throughput by 5×.

Software Engineer - Goldman Sachs

(2019-01 - 2022-05)

• Developed large-scale data integrity and deduplication algorithms, improving cross-dataset accuracy and reliability.
• Co-designed secure multi-tenancy architectures for high-sensitivity financial data, supporting robust data isolation and access control.

Master of Engineering - Scientific Computing - University of Pennsylvania (2026-08)

Bachelor of Science - Biomedical Engineering - Yale University (2010-05)