Please note: this event has passed
Abstract: This work explores deterioration in Alzheimer’s Disease using Machine Learning. It aims to contribute to an existing body of literature which addresses deterioration in dementia. This aspect of Alzheimer’s research is challenging and multi-faceted. Indeed, recent literature indicates that the progression within dementia is heterogenous, both within and between persons, and originates from disease characteristics [3], and suggests that possible risk factors should be evaluated at baseline, but also at intervals during disease progression. To the first point, our current work [1] suggests that heterogeneity exists, at the very least between the two patient groups considered in this study (details below). Such heterogeneity would indicate differing approaches may be appropriate with these groups, when selecting treatments or interventions. To the second point, this work suggests a difference in the expression of risk factors at baseline, as compared to the subject’s final visit.
In this paper exploring deterioration in Alzheimer’s Disease using Machine Learning [1], subjects were split into two datasets based on baseline diagnosis (Cognitively Normal, Mild Cognitive Impairment), with outcome of deterioration at final visit (a binomial essentially yes/no categorisation) using data from the Alzheimer’s Disease Neuroimaging Initiative ADNI (demographics, genetics, CSF, imaging, and neuropsychological testing etc). Six machine learning models, including Random Forest (RF), Support Vector Machines (SVM), Gradient Boosting Machine (GBM), Elastic Net (EN), Gaussian Processes (GP), and Classification and Regression Tree (CART), were built, and evaluated on these datasets using a nested cross-validation procedure, with the best performing models being put through repeated nested cross-validation at 100 iterations. We were able to demonstrate good predictive ability (CART: AUC = 0.88), with models predicting which of those in the cognitively normal group deteriorated and received a worse diagnosis . For the mild cognitive impairment group, we were able to achieve good predictive ability for deterioration (EN: AUC = 0.76).
Links:
[1] https://research.gold.ac.uk/id/eprint/31525/
[3] https://pubmed.ncbi.nlm.nih.gov/25312773/
Bio: Henry Musto has an MA in Psychology and a MSc in Brain Imaging Methods from University of Glasgow. After his MA and MSc studies, he conducted a research internship in predicting autism spectrum quotient (AQ) under the supervision Prof Simon Baron-Cohen at Cambridge University's Autism Research Centre.
At present Henry conducts his part-time PhD research in predicting Dementia with Machine Learning, developing his work in the Data Science & Soft Computing Lab at Goldsmiths College, University of London, under the supervision of Dr Daniel Stamate (Goldsmiths and Manchester University) and Prof Daniel Stahl (King's College London). He also works as a full time data scientist in industry.