Computer-aided diagnosis are systems that assist doctor in interpretation of medical images, and hence in making a diagnosis of an illness. Also the biomarkers that helps the diagnosis can originated from different imaging modalities, and therefore provide different possible interpretations and meanings per modality. That is why finding the optimal combination of those multimodal parameters is essential, and should lead to a higher discrimination/classification power (healthy vs condition) than using solely one imaging modality.
Our research concerns a newly discovered medical condition – Accelerated Cognitive Ageing (ACA) – in which patients suffer from a faster decline in cognitive functions as compared to cognitive decline caused by typical ageing. In other terms people with ACA tend to have an older brain, with respect to cognitive functions, than their aged-matched healthy peers. By looking at the brain with multimodal Magnetic Resonance Imaging (MRI) techniques in combination with assessment of cognitive functions, different correlations between MRI parameters and cognitive scores might arise. This potentially will allow us to detect early biomarkers and may also – if these processes can be interpreted causally- provide a model for cognitive ageing.
The ACA is currently investigated by psychologists, radiologists and researchers, in order to find neurological biomarker of such a cognitive decline. This project also aims at eventually building a reliable classification tool for computer-aided diagnosis of ACA. The project is part of the multi-disciplinary program called Neu3ca (http://neu3ca.org/), which aims at imaging, diagnosing, and treating such brain neurodegenerative disorders that affects the ageing brain.
Multiple modalities have been used to scan the brain of patients with ACA and healthy age-matched control peers. Here, we focus on MRI-based techniques including functional MRI, structural MRI, diffusion and perfusion MRI. Using functional MRI, we can extract dynamic measurement of brain functional networks. Also with structural imaging we can obtain measures of cortical thickness, or the white matter density of someone’s brain. Combining these MR parameters together with demographic information (age, education level, etc.) in a machine learning process can lead to automated diagnosis for early ACA detection. Different machine learning and classification techniques have to be investigated. A selection of the most sensible techniques —for brain MRI— that combine pattern recognition, multi-kernel learning and decision tree classification would be favored. The research question is the following. What features are best descriptors of ACA and how to combine them to obtain a significant high classification performances?
If classifiers are trained with the right features and yield to high accuracy in classification, such tool can be utilized for computer-aided diagnostic decisions for ACA in epilepsy.