My research topics primarily focus on the automatic segmentation of multidimensional medical data, with applications in pathology understanding, follow-up, and digital twin development. These topics are part of the team Myriad.
A significant part of my work has been dedicated to Mean-shift, which is an exciting approach that does not require assumptions about the data. Mean-shift can be derived in a knowledge discovery framework. Extending this framework to spatiotemporal data, scale, and space selections, with the integration of a few prior, constituted my main research. Such methods are motivated by medical challenges ( Multiple sclerosis, stroke, cardiac segmentation…. ).
Now, I am focusing on approaches based on deep learning for segmentation, filtering, and localization/detection tasks that can provide better and more robust results than many conventional approaches. The main challenges are the use of semi- and weakly-supervised methods, dealing with experts’ disagreements, explainability, confidence, and trying to train in a playful manner (selection of needed data, continuous learning, …). The applications of such segmentations are: pathologies understanding and quantification, longitudinal analysis (disease evolution, clinical care), and more and more for numerical simulations (CFD, failure load…: digital twin).
For image processing application development, my background is on ITK and OpenCV, with QTCreator and CMake for C++, under Windows and Linux. I use more and more Python with conda and really appreciate the jupyter lab environment (debuging and contextual help !).
For deep learning, I mostly use Python with Keras/Tensorflow (using conda), or MONAI . For Yolo, I switch from AlexeyAB Darknet lib to Ultralytics one.