Filima Patrick

FILIMA PATRICK

Research Software Engineer & Technical Product Lead

HomeAboutExperienceResearchPublications
© 2026 Filima Patrick
Back to Projects & Experience
Translational Medical AI & Clinical Protocol OptimizationActive Clinical Translational Study / Project 7
License: MIT

DeepAccess-MRI: AI-Assisted MRI Protocol Optimization via Missing Sequence Reconstruction

Clinically Validated Transfer Learning & Missing Sequence Synthesis for Shorter, Affordable Neuroimaging Protocols in Nigeria

Python 3.11+PyTorchMONAIDiffusion ModelscGANsU-NetSimpleITKMonte Carlo DropoutRadiologist Audit
Core Research Ecosystem Vision

Developing and clinically validating a transfer-learning framework that synthesizes missing structural MRI sequences (e.g. T1 -> FLAIR/T2) from heterogeneous Nigerian clinical scans, enabling shorter, non-inferior acquisition protocols without compromising diagnostic confidence.

5 Translational Research Objectives

Objective 1: Transfer Learning & Domain Adaptation

Fine-tuning international research-grade pre-trained weights (BraTS, IXI, HCP) on local Nigerian clinical datasets.

Objective 2: Radiologist Non-Inferiority

Evaluating whether AI-assisted reduced protocols match full physical acquisition protocols in diagnostic accuracy.

Objective 3: Pathology Preservation & Safety

Auditing that synthetic sequences preserve disease-specific pathology (ventriculomegaly, lesions, atrophy) without hallucinations.

Objective 4: Quality & Noise Thresholding

Locating exact limits of clinical noise, motion blur, and SNR where sequence reconstruction remains safe.

Objective 5: Multi-Center Generalization

Validating zero-shot transferability across LifeBridge, UPTH, and RSUTH hospital networks.

DeepAccess-MRI Software Prototype ModuleFunctional Clinical Deliverable
Clinical Quality AssessmentAutomated evaluation of input scan quality against minimum safety thresholds.
Sequence Inventory AuditAutomatic identification of present and missing MRI sequence modalities (T1, T2, FLAIR).
Missing Sequence ReconstructionClinically validated synthesis of missing target sequences using transfer-learned generators.
Voxel-Wise Uncertainty MappingMonte Carlo Dropout spatial confidence maps highlighting low-confidence reconstruction zones.
Protocol Optimization Decision SupportActionable recommendations for reduced physical scan times without diagnostic confidence loss.
Planned Peer-Reviewed Publications & Doctoral Outputs
📄Paper 1: Characterizing Real-World Nigerian Clinical Brain MRI Data for AI Development
📄Paper 2: Transfer Learning and Domain Adaptation for Multimodal MRI Sequence Synthesis using Clinical Data
📄Paper 3: Quantifying the Impact of Motion, Noise, and Resolution Degradation on Deep Learning Synthesis Reliability
📄Paper 4: Pathology Preservation, Hallucination Audits, and Radiologist Non-Inferiority Assessment of AI-Reconstructed MRI Sequences
📄Paper 5: AI-Assisted Protocol Optimization: Clinical and Economic Evaluation of Shortened MRI Examinations in Nigerian Hospitals

Execution Commands & Pipeline Workflow

# 1. Train MONAI 3D DenseNet-121 Benchmark:
python src/models/cnn.py --model densenet121 --task task4 --epochs 50 --batch_size 4 --out_dir ./checkpoints/densenet
# 2. Evaluate Degradation Robustness & RRI Metrics:
python src/evaluation/metrics.py --evaluate_robustness --model_dir ./checkpoints/ --out_dir ./results/
# 3. Generate Grad-CAM Explainability Heatmaps under Perturbation:
python src/evaluation/explain.py --model_path ./checkpoints/densenet/best.pth --method gradcam --out_dir ./results/explanations/