Nicolas Gaggion

Nicolas Gaggion

PhD in Engineering: Signals, systems and artificial intelligence. Machine Learning Engineer.

Research Lines

I focus on two primary lines of research that explore the intersection of deep learning, graph-based representations, and image analysis.

  • High Throughput Phenotyping of Plant Roots Using Deep Learning:
    During my Computing Engineer thesis, I collaborated with Federico Ariel from the Instituto de Agrobiotecnología del Litoral (IAL) and researchers from the Institute of Plant Sciences Paris Saclay (IPS2), a laboratory where I conducted research during 2022. Together, we developed a deep learning-based methodology for high throughput phenotyping of plant roots from time-series images.
    Currently, our objective is to operationalize this research, making it accessible for use in plant biology labs worldwide.
    As part of this project, we organized a root segmentation challenge during the CVPPA workshop at ICCV ‘21.

  • Image-to-Graph Extraction via Convolutional and Graph Neural Networks for Medical Imaging:
    My PhD thesis concentrates on the extraction of graph structures from images in the medical imaging domain. I am specifically exploring this approach for image segmentation, using graph representations of the anatomical structure rather than traditional pixel-level segmentation.
    In July 2022, I visited the Center for Computational Imaging & Simulation Technologies in Biomedicine, CISTIB, under the supervision of Prof. Alex Frangi. Our collaborative research involves automatic mesh extraction from cardiac MRI images using UK Biobank data.

Although these research lines appear distinct, both revolve around the concept of extracting graph structures from images. In the plant root phenotyping project, a combination of deep learning-based segmentation and conventional graph construction approaches results in a unique graph representation of the root structure. In the medical imaging context, I am developing end-to-end trainable models for anatomical segmentation using graph neural networks. These endeavors highlight the synergy between deep learning and graph-based representations in the realm of image analysis.