Categorize, cluster & classify: The 3C strategy applied to Alzheimer's disease as a case study

Health informatics is facing many challenges these days, in analysing current medical data and especially hospital data towards understanding disease mechanisms, predicting the course of a disease or assist in targeting potential therapeutic options. Alongside the promises, many challenges emerge. Among the major ones we identify: current diagnosis criteria that are too vague to capture disease manifestation; the irrelevance of personalized medicine when only heterogeneous classes of patients are available, and how to properly process big data to avoid false claims. We offer a 3C strategy that starts from the medical knowledge, categorizing the available set of features into three types: the patients' assigned disease diagnosis, clinical measurements and potential biological markers, proceeds to an unsupervised learning process targeted to create new disease diagnosis classes, and finally, classifying the newly proposed diagnosis classes utilizing the potential biological markers. In order to allow the evaluation and comparison of different algorithmic components of the 3C strategy a simulation model was built and put to use. Our strategy, developed as part of the medical informatics work package at the EU Human Brain flagship Project strives to connect between potential biomarkers, and more homogeneous classes of disease manifestation that are expressed by meaningful features. We demonstrate this strategy using data from the Alzheimer's Disease Neuroimaging Initiative cohort (ADNI).