Personalized modeling for medical decision support
| aut.embargo | No | en_NZ |
| aut.thirdpc.contains | No | en_NZ |
| aut.thirdpc.permission | No | en_NZ |
| aut.thirdpc.removed | No | en_NZ |
| dc.contributor.advisor | Kasabov, Nikola | |
| dc.contributor.advisor | Feigin, Valery | |
| dc.contributor.author | Liang, Wen | |
| dc.date.accessioned | 2013-08-18T23:03:29Z | |
| dc.date.available | 2013-08-18T23:03:29Z | |
| dc.date.copyright | 2013 | |
| dc.date.created | 2013 | |
| dc.date.issued | 2013 | |
| dc.date.updated | 2013-08-18T20:18:08Z | |
| dc.description.abstract | Personalized modeling is an emerging approach, in which a model is created for every new input vector of the problem space based on its nearest neighbors using transductive reasoning (Kasabov, 2007c; Vapnik, 1998). The underlying philosophy of this approach when applied to medicine is that each patient is an individual. Therefore, each patient requires and deserves a personalized treatment model that predicts the best possible outcomes for the patient. This study proposes a novel integrated evolving framework and system for personalized modeling (evoPM); an extension of a model proposed by Kasabov and Hu (Kasabov & Hu, 2011). By allowing users to select the most important features, optimize nearest neighbors and model parameters, the model provides higher accuracy and personalized knowledge than global and local modeling approaches. The evoPM creates a personalized model for each test sample with unique optimal sets of features, neighborhood and model parameters. In addition, the system keeps evolving and is adaptable to any new incoming data vectors. The already created personalized model can be further evolved on new data entering in the neighborhood. Currently, the amount of available spatio-temporal data (STD) is growing exponentially, thus suitable techniques to e ectively and e ciently analyze and process this vast quantity information are urgently needed. Evolving spiking neural networks (eSNN), an extension of spiking neural networks (SNN), is an emerging computational technique for STD analysis. Evolving SNNs learn STD by rst converting temporal changes in the input variables into spike trains, then applying learning procedures to map spatio-temporal patterns detected in the data into temporal spiking activity of spatially located neurons. This study introduces two recently proposed methods for spatio-temporal pattern recognition, the extended eSNN framework (EESNN) (Hamed, Kasabov, Shamsuddin, Widiputra & Dhoble, 2011) and the recurrent network reservoir structure of eSNN (reSNN) using liquid state machine (LSM) (Schliebs, Hamed & Kasabov, 2011). Both methods are the rst time applied to evaluate the spatio-temporal weather and stroke occurrence data as a case study. The evoPM is applied as a classi er to learn the responses from the reSNN model. The novel evoPM framework and system brings several advances over existing per sonalized modeling methods. These are summarized below: The integrated evolving personalized modeling system is developed based on an emerging novel technology namely eSNN; A recently developed population-based heuristic optimization approach called gravitational search algorithm (GSA) is applied to improve the robustness and general disability of feature selection, neighborhood, model and its parameters optimization for classification, diagnostic and prognostic problems; The standard diseased classification system is replaced by personalized risk evaluation. The evoPM system and framework is novel applied to stroke data as case studies. The novel method is validated on several benchmark cancer gene expression datasets and stroke data. The model outputs are compared with those of traditional global, local and personalized modeling methods. The results of all studies show that evoPM performs consistently better than the traditional methods. In particular, it develops more useful knowledge discovery for medical decision support for cancer diagnosis and prognosis due to it selects the optimal sets of genes and disease classification parameters for each individual patient. | en_NZ |
| dc.identifier.uri | https://hdl.handle.net/10292/5607 | |
| dc.language.iso | en | en_NZ |
| dc.publisher | Auckland University of Technology | |
| dc.rights.accessrights | OpenAccess | |
| dc.subject | Personalized modeling | en_NZ |
| dc.subject | Stroke | en_NZ |
| dc.subject | Medical decision support | en_NZ |
| dc.title | Personalized modeling for medical decision support | en_NZ |
| dc.type | Thesis | |
| thesis.degree.discipline | ||
| thesis.degree.grantor | Auckland University of Technology | |
| thesis.degree.level | Doctoral Theses | |
| thesis.degree.name | Doctor of Philosophy | en_NZ |