Background: Brain image analysis has become one of the main and essential tools for diagnosis, monitoring and prognosis of brain diseases including multiple sclerosis and stroke. Despite the advances done, the impact on treatment and on care of patients is limited due to the same disease’s complexity and lack of precision.
Main hypothesis: The introduction of deep learning techniques to automatically extract and learn structural information from brain images have improved state-of-the-art performances of many brain image biomarkers. However, although relevant clinical studies have highlighted the importance of these models, the current available tools are not robust and accurate enough to provide predictive models usable in clinical practice. Our hypothesis in this project is that by using innovative methods such as the quantification of brain aging, the improvement of longitudinal change estimates, and the integration of novel semi-supervised, self-supervised and continuous learning strategies, could boost the development of imaging predictive models, that together with the fusion with clinical and patient measures will allow the creation of novel and more precise predictive models of brain diseases evolution.
Objective: The main goal of this project is to investigate, develop and validate novel and advanced image analysis and deep learning solutions to improve current technology in brain imaging. The project proposes to push beyond the state of the art in different research areas: 1) the development of models for brain age estimation from MRI images that could be used as a biomarker for neurodegenerative diseases; 2) the development of transversal and longitudinal models that could provide more robust change detections; 3) the development of tools for self-supervised and semi-supervised learning, aiming to obtain more robust and generalisable models and to transfer biomarkers and predictive models to different tasks, providing also the capability to perform incremental and continuous learning through the addition of new data; 4) the development of pioneer predictive models fusing the aging biomarker with other relevant clinical data for multiple sclerosis patients; and 5) the development of novel deep learning models for stroke patients to predict patient recanalization (TICI score) from basal CTA images, and to predict hemorrhagic expansion using basal CT images.
Expected results: The main outputs of this project will be: 1) automated tools tested in different MRI and CT scanners of the collaborating hospitals; and 2) predictive models to improve assessment of diseases evolution. Advances in this research line may have an effect in patient prognosis, but also in economic terms given the associated costs of ineffective treatments, which suggests that new research on the field can be extremely beneficial to move towards individualised therapeutic approaches (patient stratification).
Collaborators:
Hospital Dr. Josep Trueta de Girona (IDIBGI Institute)
Hospital Vall d’Hebron de Barcelona (VHIR Institute)
Neurological diseases, magnetic ressonance images, computed tomography images, new technologies, deep learning, imaging biomarkers, predictive models
