Wanpracha Chaovalitwongse Rutgers University New Brunswick
CAREER: NOVEL OPTIMIZATION METHODS FOR COOPERATIVE DATA MINING WITH HEALTH-CARE AND BIOTECHNOLOGY APPLICATIONS
There is an urgent need to advance and apply quantitative and qualitative approaches to the study of epilepsy and brain disorders. As uncontrolled epilepsy poses a significant burden to society due to as- sociated healthcare cost, this project is aimed at the development of an automated seizure prediction system and brain abnormal activity classifier. To achieve this goal, optimization-based data mining (DM) approaches will be developed to quantitatively analyze the brain activity through electroen- cephalogram (EEG) data. The proposed DM techniques will excavate hidden patterns/relationships in EEGs, which will give a greater understanding of brain functions (as well as other complex sys- tems) from a system perspective. Specifically, a new DM paradigm for the seizure prediction and brain activity classification will be developed based on novel optimization-based DM techniques for feature selection, clustering, and classification. The proposed research will contribute to the computer science, engineering and medical communities along the following four lines: (1) the development of novel mathematical models and optimization techniques for DM problems and time series analysis, (2) the implementation of statistical techniques to detect patterns from selected features/clusters for predicting seizures and classifying normal and epileptic EEG activity, (3) the utility of detection theory and the experimental designs to assess and validate the efficacy, robustness, and uncertainty of the proposed DM paradigm as well as fine-tune the optimal parameter setting, (4) the extension of the fundamental research findings in optimization and DM to other cross-disciplinary research, which will constitute a new avenue of research in optimization-based DM and time series analysis. The proposed research is very crucial to decision making processes in real world problems. Success of this research will advance the state-of-the-art in the field of optimization in DM, and have a greatly significant impact on medical research. The research scope in this proposal touches upon several emerging optimization and DM problems, which are driven by ever growing computational power. The proposed research has shown a broad impact on many research fields including computer science, operations research, computational biology, and logistics. The scope of this project itself will broaden opportunities and enable the participation of all citizens women and men, underrep- resented minorities, and especically persons disabled by epilepsy. Success of this proposal in seizure prediction research will relieve the anguish from this life-threatening disease and improve the life quality of at least 2 million Americans (14 millions worldwide), who are currently suffering from epilepsy regardless of race, age, or gender.
The outcome of this NSF CAREER proposal can be categorized into 2 streams: computational foundation and real life applications. I developed a number of mathematical models and algorithms for identifying patterns in complex data. They were designed to reveal important characteristics of the data that can be used to infer the group or class of each of the data samples. I constructed a new computational framework for these mathematical models and algorithms, and tested it in many real life problems. The first problem deals with brain activity signals via electroencephalogram (EEG). I used my framewotk to identify patterns in EEG signals that are associated with epileptic seizures, and employed an online monitoring technique to predict impending seizures. I also used my framework to predict, from EEG signals, if a user will make mistakes in his/her operation - specifically, numerical typing. In addition, I used my framework to detect, from EEG signals, when a user was having a high cognitive load from the task that he/she was performing. Another application of my framework is to probabilisticly identify targets/objects from a continuous 3D images. This application is very relevant in airforce operations, e.g., unmanned drones. The mathematical models developed in this project were also investigated and applied to logistics optimization problems as well as computational biology problems.
