Background: Multiple sclerosis (MS) is a chronic inflammatory disease of the central nervous system, with a progressive neurodegenerative course. MS causes disability with both physical and cognitive deficits, which have a huge influence on personal, social and working life of patients and their families. There is no cure for the disease, but existing immunomodulatory treatments reduce the number of clinical relapses and are helpful to decrease the chances of disability progression. In this regard, predicting the next stage of the disease is crucial to select the best treatment to interfere in the progression of the disease. Despite the advances in the development of image analysis techniques using magnetic resonance imaging (MRI), the impact on treatment and on care for MS patients is limited, due to the same disease complexity, imaging restrictions and the inherent lack of precision of computerised tools and models.
Main hypothesis: Novel deep learning techniques have recently shown a significant increase in the accuracy on different computer vision tasks, mostly by their capability to extract richer features and build more deeper models. Our hypothesis is that deep predictive models of the disease evolution can be proposed using not only those new rich image descriptors but also fusing them with available clinical information. Advances in this line may have a direct effect not only 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.
Objective: The main goal of the EVOLUTION project is to investigate, develop and validate new automated computerised tools to extract MRI biomarkers with the aim of computing MS predictive models that can be applicable in clinical practice. This novel paradigm will provide the basis for improving the prognosis and monitoring of MS, introducing objectivity, and simplifying the everyday practice for radiologists and neurologists. From the technological point of view, the project pushes the state of the art in various areas, investigating: i) the use of deep learning techniques to improve the current MRI brain biomarkers for MS such as brain lesions, brain atrophy, and cortical and subcortical grey matter; ii) novel solutions to deal with the generalisation of these models to different image domains (i.e. different MRI scanners or image resolutions) without requiring large amount of training data; iii) the development of pioneer predictive models fusing MRI information with other relevant clinical data such as psychological, disability and genetic data to infer the probability of future events such as the EDSS disease indicator or the probability of the effectivity of a particular drug. At early stages these predictive models could also serve to design a MRI screening protocol to predict the risk of developing the disease.
Expected results: This project will generate the following relevant ouptuts: i) automated and generalisable tools tested in three different MRI scanner machines of the collaborating hospitals (Siemens, Philips and General Electric) including also different magnetic field strenghts (1.5T and 3T images); ii) predictive models to improve assessment of MS evolution and patient stratification. Such tools would be ready for application to other neurodegenerative diseases, helping to determine and quantify disease evolution.
