Today, patients' data often include a large amount of structured information, such as neuroimaging data, neuropsychological tests results, demographical variables, etc. However, human beings cannot analyze so much information without the help from modern data mining and machine learning methods. Given these rich and diverse information sources, computer scientists can develop computerized methods to discover hidden knowledge, patterns that may help clinicians gain deep insights into diseases. Further, computer-aided diagnosis (CAD) is also a tool to assist clinicians make critical diagnosis decisions. Therefore, the goal of the study is to devise computer algorithms that can uncover useful patterns, which can further be used to build models for prediction. As a result, both descriptive knowledge and predictive models can be attained.
Furthermore, it is a common practice to combine different information sources to reach a more trustful result. Multiple kernel learning, majority voting and stacking are candidates in this regard, to name only a few. A stacked multi-view learning is devised, making decision based on base and meta level learning. The base level collects decisions provided from base level classifiers and the meta level classifier tries to learn the correlation between correct decision and decision behaviors among base classifiers. The proposed method yields better result than using any individual information source alone. Therefore, it is encouraged to meaningfully combine information sources