Keywords: machine learning, neuroimaging, precision medicine, stroke, self-attention Internship Duration: 30/11/-1 - 30/11/-1
Head of the hosting team: Nazim Agoulmine
Website: Click here
Address of the host laboratory: IBISC EA4526 Team Vincent VIGNERON 40 rue du Pelvoux 91020 Courcouronnes France
Supervisor: Vincent VIGNERONE-mail: vincent.vigneron@univ-evry.fr Phone: +33663568760
In the event of a stroke, an accurate diagnosis must be made as soon as possible by a highly qualified specialist. This diagnosis involves the visual interpretation of a brain MRI to decide on endovascular recanalization treatment. The doctor must quickly choose a treatment option (single or double thrombolysis, thrombectomy) by a quick overview of the images, without the possibility of in-depth evaluation, in a context where every untreated minute increases the risk of death or disability. Our team developed a diagnostic aid algorithm that meets these requirements of speed and reliability for hyperacute ischemic strokes (<3 hours). It is the first comprehensive automatic tool for simultaneous segmentation of lesions. The segmentation of the lesion is today of the same level of performance as a segmentation carried out by a neurologist. Ultimately, the technique should provide more information than the human eye regarding the texture of the lesion and its visibility. Artificial intelligence will also be able to provide numerical values of relevant parameters instead of the approximations currently used. This technology is intended for health professionals (Neuro-Vascular Unit and emergency services), regardless of their experience or expertise, and will ensure a diagnosis in the event of the absence of an expert on-site (telemedicine). The gain in assistance is even more important for a minority of patients for whom the medical decision is difficult. Some important questions remain: is the segmentation robust to most MRI equipments? Can we evaluate the sensitivity of the segmentation when some modalities are missing? which prior information can be used to remove false positives? The objectives of the internship is to design new algorithms and validate them on a large patient database and to integrate the model into clinical application software. Learning by neural networks inspired by ladder networks or regime networks or adversary auto-encoding, curricular model, etc. will be privileged in this project. The expected solution will better characterize stroke by associating multiple weak signals with the definition of pathology, which can therefore lead to replacing complex clinical scales (CPSSS, NIHSS, LAMS, VAN, etc.), which are tedious to calculate in current practice. It will improve the quality of reading of current images, for example, by performing analyses that are not currently carried out because they take too long to execute manually such as the volumetric measurement of the lesion, the extraction of textures, etc. This solution will be able to determine what information in the image implies certain treatments leading to better results for patients. AI can help the radiologists prioritize urgent cases by determining in advance which imaging tests to assess first.
Deep neural network models, SPM toolbow for Python and statistical tools Programming skills: Python or C / C ++. A practice of Tensorflow and Pytorch would be a plus.
[1] Ikram Brahim. Deep Learning Methods for MRI Brain Tumor Segmentation: a comparative study. In 9th IEEE International Conference on Image Processing Theory, Tools and Applications (IPTA 2019), Istanbul, Turkey, November 2019. [2] Liang Chen. Fully automatic acute ischemic lesion segmentation in dwi using convolutional neural networks. NeuroImage: Clinical, 15:633 – 643, 2017. [4] J. Egger, Graph-based tracking method for aortic thrombus segmentation. 4th European Conference of the International Federation for Medical and Biological Engineering, pages 584–587, Berlin, Heidelberg, 2009. [6] Oskar Maier, les 2015 - a public evaluation Medical Image Analysis, 35, 07 2016.
