Hector Lise de Moura

Predictive Modeling for Post-Surgical Osteoarthritis Risk

Developed a predictive model to identify patients at high risk of developing post-traumatic osteoarthritis after ACL reconstruction surgery. Integrated and analyzed multimodal data, including MRI scans, clinical reports, demographic data, and functional patient metrics, to identify leading indicators of disease progression. The project involved identifying complex patterns and risk factors in medical datasets to forecast outcomes — a process directly analogous to identifying indicators of fraud or financial risk in transactional data.

Technologies: Python, Machine Learning, Medical Imaging, Data Integration, Risk Assessment

Impact: Early risk identification, improved patient outcomes, data-driven healthcare decisions

Advanced Ultrasound Signal Processing for Equipment Inspections (AUSPEX)

Led the development of software and inspection tools for Petróleo Brasileiro S.A. (Petrobras), creating methods to increase the reliability of submarine pipeline inspections while reducing operational costs.

Technologies: Python, PyTorch, Qt5, NumPy, SciPy, CUDA, WebAssembly, Git, CI/CD

Impact: Enhanced submarine inspection reliability, reduced costs, improved safety protocols

View Project Page →

Knee Cartilage and Bone Segmentation

Developed a pipeline for image reconstruction and automatic segmentation of MRI images of the knee, eliminating the need for manual segmentation by radiologists. This method achieved 4th place out of 87 competing teams at the 2022 MICCAI K2S Challenge.

Technologies: Python, PyTorch, SLURM, Medical Imaging, Deep Learning

Impact: Reduced manual segmentation time by 90%, improved process reliability

Multi-modal Imaging Super-resolution for Sodium MRI

Developed a fusion technique that combines low-resolution images with higher-resolution images from various MRI contrasts to enhance the resolution of sodium MRI. The method reduced inference time from ~616s using PLS methods to ~1.2s using a dual-encoder Y-net architecture.

Technologies: Python, PyTorch, SLURM, Multi-modal Imaging, Neural Networks

Impact: 500x speed improvement, maintained image quality, enabled real-time processing

View on GitHub →

Simultaneous Bilateral T1, T2, and T1ρ Relaxation Mapping of Hip Joint With 3D‐MRI Fingerprinting

Anmol Monga, Hector Lise de Moura, Marcelo VW Zibetti, Thomas Youm, Jonathan Samuels, Ravinder R Regatte

Three‐dimensional MR fingerprinting (3D‐MRF) has been increasingly used to assess cartilage degeneration, particularly in the knee joint, by looking into multiple relaxation parameters. A comparable 3D‐MRF approach can be adapted to assess cartilage degeneration for the hip joint, with changes to accommodate specific challenges of hip joint imaging.

View Publication →

Fine‐Tuning Deep Learning Model for Quantitative Knee Joint Mapping With MR Fingerprinting and Its Comparison to Dictionary Matching Method

Xiaoxia Zhang, Hector L De Moura, Anmol Monga, Marcelo VW Zibetti, Ravinder R Regatte

Deep learning (DL)–based methods for quantification mapping in MRF overcome the memory constraints and offer faster processing compared to the conventional dictionary matching (DM) method.In this study, we investigate the impact of training parameter choices on NN performance and compare the fine‐tuned NN with DM for multiparametric mapping in MRF.

View Publication →

Optimized MR pulse sequence for high‐resolution brain 3D‐T1ρ mapping with weighted spin‐lock acquisitions

Marcelo VW Zibetti, Rajiv Menon, Hector L De Moura, Mahesh B Keerthivasan, Ravinder R Regatte

To implement and evaluate the feasibility of brain spin–lattice relaxation in the rotating frame (T1ρ) mapping using a novel optimized pulse sequence that incorporates weighted spin‐lock acquisitions, enabling high‐resolution three‐dimensional (3D) mapping.

View Publication →

Feasibility of a UTE Stack‐of‐Spirals Sequence for Biexponential T1ρ Mapping of Whole Knee Joint

Hector L de Moura, Mahesh B Keerthivasan, Marcelo VW Zibetti, Pan Su, Michael J Alaia, Ravinder Regatte

This study investigates the feasibility of a UTE Stack-of-Spirals sequence for biexponential T1ρ mapping of the whole knee joint, aiming to improve the accuracy and efficiency of knee joint imaging.

View Publication →

MR electrical properties mapping using vision transformers and canny edge detectors

Ilias I Giannakopoulos, Giuseppe Carluccio, Mahesh B Keerthivasan, Gregor Koerzdoerfer, Karthik Lakshmanan, Hector L De Moura, José E Cruz Serrallés, Riccardo Lattanzi

In this paper we developed a 3D vision transformer‐based neural network to reconstruct electrical properties (EP) from magnetic resonance measurements.

View Publication →