Bridging AI and science to transform research.

Mission:
Developing multimodal AI models to link MRI, histology, and spatial transcriptomics for non-invasive ovarian cancer characterization. Goal: Predict tumor biology, immune landscapes, and treatment response from imaging.

Key Work:

• Multimodal Coregistration: Align 3T/9.4T MRI, histology, and transcriptomics at voxel-level (ANTs, Coherent Point Drift).
• Feature Extraction: Radiomics (PyRadiomics), pathomics (QuPath), transcriptomics (scuttle/scran).
• AI Models: U-Net (segmentation), CNNs (fusion), spatial transformers (context). Loss functions: Hausdorff (best MCC: 0.65), Tversky, Dice.
• Clustering: BayesSpace, SNNGraph for spatial gene expression patterns.

Progress:

• 12 HGSOC samples (100GB+), 9.4T MRI, 10X Visium transcriptomics, H&E slides.
• Segmentation models: 98% specificity, 65% MCC (Hausdorff Loss).
• Spatial clustering: 3 distinct tumor habitats identified (immune hot/cold, stroma-rich).

Impact:

• First spatially aligned multimodal dataset for HGSOC.
• AI tools to predict immune phenotypes and therapy response from MRI.
• Open-source pipelines for radiogenomics and precision oncology.

Next: Validate models on external cohorts; deploy clinical decision support tools.

Mission:
Designed and deployed AI-driven solutions for scientific imaging (microscopy, histopathology, agronomy), bridging research and operational tools. Developed end-to-end pipelines—from data preprocessing to model deployment—enabling partners to automate analysis and accelerate discovery.

Key Projects:

• HixLoop: Built a modular AI platform for object detection/segmentation in scientific images, cutting manual annotation effort by 80% and improving model precision to 98%. (FastAPI, PyTorch, Docker, Streamlit)
• Abbelight: Developed CNN models for super-resolution microscopy, boosting recall/precision by 15–20% and delivering a cross-platform pipeline. (PyTorch, MLflow, GPU optimization)
• ULTIM: Automated transcription site detection in microscopy, reducing analysis time from hours to minutes. (Detectron2, Docker, Streamlit)
• Tools: Created GPUBalancer (GPU queue system), CIRToolbox (time-tracking with voice AI), and solar shadow simulators for agronomic research.

Skills:
AI/ML: PyTorch, TensorFlow, Keras, Detectron2
DevOps: Docker, FastAPI, GPU clusters
Collaboration: Git, Notion, Agile

Impact:

• Delivered 3 core platforms now used daily by BionomeeX and partners (IRCM, Abbelight).
• Set standards for modular AI pipelines and Docker-based deployment, reducing project time by 60%.
• Contributed to publication-ready tools for oncology and microscopy research.

Next: Applying this expertise in my PhD at IRCM, developing multi-modal AI for ovarian cancer detection.