Cognitive impairment in neurodegenerative diseases: insights from computational neuroimaging
MetadataShow full item record
Background: Cognitive impairment is a very common problem, especially in older individuals with major impact on quality of life, daily functioning, and healthcare. Its importance is expected to increase due to the demographic changes. Neuroimaging is a rapidly developing field of neuroscience that provides an opportunity to study brain mechanisms of cognitive impairment in vivo, which may help in the development of new biomarkers and treatment strategies. The application of advanced image processing to neuroimaging offers the potential for diagnostically relevant analysis techniques, in particular for magnetic resonance imaging (MRI).
Aim: The primary aims of the project were to investigate brain mechanisms of cognitive impairment in neurodegenerative diseases using computational neuroimaging approaches and to assess their potential applicability in clinical practice for detection, prediction and differential diagnosis of cognitive impairment in the elderly.
Objectives: 1) To investigate brain changes underlying cognitive impairment in neurodegenerative diseases (Alzheimer’s, Lewy body dementia and Parkinson’s disease). 2) To assess the applicability of pattern recognition techniques for: a) Differential diagnosis of cognitive impairment b) Prediction of further cognitive deterioration in patients with mild cognitive impairment; 3) To investigate problems associated with implementation of computeraided image-based tools for detection, prediction and differential diagnosis of cognitive impairment.
Methods: Five datasets of clinical and imaging data were used, including two large-scale databases of Alzheimer’s disease (ADNI and AddNeuroMed). In the papers I-II, Alzheimer’s disease was diagnosed according to the NINCDSADRDA criteria. Dementia with Lewy bodies (paper I) was diagnosed according to the revised consensus criteria (1) Image post-processing steps were performed within the surface- (papers I-III) and voxel-based (paper IV) frameworks using the Freesurfer and SPM8, respectively. Mass-univariate (papers III, IV) and multivariate (papers I, II and IV) approaches were used. In the paper IV, an automated quantitative metaanalysis was also performed using the Neurosynth software.
Results: Papers I-II: Optimizing image preprocessing and data analysis pipeline, we found that it is possible to develop a computer-aided tool for detection (Sensitivity/Specificity = 88.6%/92.0%), prediction (Sensitivity/Specificity = 83.3%/81.3%) and differential diagnosis (AD/DLB overall classification accuracy = 83.9%) of degenerative diseases with good between-cohort robustness if imaging and clinical protocols are carefully aligned. For the morphometric data, the use of disease-specific brain parcellation schemes resulted in equivalent performance compared to normalized raw high-dimensional input, but required substantially lesser tuning time and computation/memory resources. Better accuracy of the models can be achieved by adding more biomarkers (e.g., ApoE genotype), demographics, and improved disease verification strategies (e.g., post-mortem diagnosis) for the data used as a training material for the classifiers.
The next two papers were focused on neural correlates of cognitive impairment in PD that had to be investigated prior considering them within the framework of computer-aided diagnosis.
Papers III-IV: We found that Parkinson’s-related cognitive impairment affecting multiple domains is associated with temporo-parietal and superior frontal thinning. On a large-scale network level, better executive performance in PD is associated with increased dorsal fronto-parietal cortical processing and inhibited subcortical and primary sensory involvement when the subject is at resting state. This pattern is positively influenced by the relative preservation of nigrostriatal dopaminergic function. The pattern associated with better memory performance favors prefronto-limbic processing, and does not reveal associations with presynaptic striatal dopamine function.
Conclusions: Cognitive impairment in the elderly has different brain profiles depending on the predominant neurodegenerative pathology and cognitive functions affected. With the use of automated computer-aided tools and advanced image processing techniques, Alzheimer’s disease can be robustly identified, predicted two years before the actual dementia onset and differentiated from dementia with Lewy bodies. After certain modifications and adaptations for clinicians, the models can be successfully incorporated into medical decision-support systems and be evaluated in subsequent diagnostic clinical trials. The identified brain structural and functional profile associated with Parkinson’srelated cognitive impairment is also robust and, holding strong diagnostic potential, must be detectable using computer-aided systems of similar design, the development of which is the matter of our future research. The development and future elaboration of clinically realistic computer-aided systems for the diagnosis of neurodegenerative diseases is an important topic for future research.