Due to scientific and technological advancements, investigations in modern medicine are producing more measurement data than ever before. Since a large amount of information exists, and it is also being produced at ever-increasing rates, no single person can digest all current knowledge of diseases. Data collected from large patient cohorts may contain valuable knowledge of diseases, which could be useful to clinicians when making diagnoses or choosing treatments. Making use of the large volumes of data in clinical decision-making requires ancillary help from information technologies, but such systems have not yet become widely available. This thesis addresses the challenge by proposing a computer-based decision support method that is suited to clinical use. This thesis presents the Disease State Index (DSI), a supervised machine learning method intended for the analysis of patient data. The DSI comprehensively compares patient data with previously diagnosed cases with or without a disease. Based on this comparison, the method provides an estimate of the state of disease progression in the patient. Interpreting the DSI is made possible by its visual counterpart, the Disease State Fingerprint (DSF), which allows domain experts to gain a comprehensive view of patient data and the state of the disease at a quick glance. In the design and development of these methods, both performance and applicability in clinical use were taken into account equally. Alzheimer's disease (AD) is a slowly progressing neurodegenerative disease and one of the largest social and economic burdens in the world today, and it will continue to be so in the future. Studies with large patient cohorts have significantly improved our knowledge of AD during the last decade. This information should be made extensively available at memory clinics to maximize the benefits for diagnostics and treatment of the disease. The DSI and DSF methods proposed in this thesis were studied in the early diagnosis of AD and as a measure of disease progression in six original publications. The methods themselves and their implementation within a clinical decision support system, the PredictAD tool, were quantitatively evaluated with regard to their performance and potential benefits in clinical use. The results show that the methods and clinical decision support tool based on these methods can be used to follow disease progression objectively and provide earlier diagnoses of AD. These, in turn, could improve treatment efficacy due to earlier interventions and make drug trials more efficient by allowing better patient selection.
|Award date||9 May 2014|
|Place of Publication||Espoo|
|Publication status||Published - 2013|
|MoE publication type||G5 Doctoral dissertation (article)|
- supervised learning
- data visualization
- clinical decision support systems