Predicting Cardiovascular Risks in Adults

One of the major cardiovascular diseases is coronary heart disease (CHD). It is reported to be a major cause of morbidity and death in adults through heart attacks and acute myocardial infarctions (AMI). CHD is a condition in which plaque builds up inside the coronary arteries, i.e., atherosclerosis. Atherosclerosis is the disease process that begins in childhood and eventually results in clinical events later in life. The factors that determine the development and progression of coronary artery disease are in large part established; however, the causes are very closely related with the risk factors present in the beginning of youth. Early detection of risks will help in designing effective treatments targeted towards the youth in order to prevent cardiovascular events in adulthood and to dramatically reduce the costs associated with cardiovascaular dieases. Our major contribution is to demonstrate the impact of machine learning on CHD research. We show that relationships between the measured risk factors and the development of advance CAD lesions and overall plaque burden can automatically be extracted and understood. As the cohort ages and sufficient clinical events occur, this work will allow us to apply these relationships to clinical events such as AMI and heart failure. Specifically, we propose using the longitudinal data collected from the Coronary Artery Risk Developments in Young Adults (CARDIA) study. This longitudinal study, started in 1985-86, measured risk factors in different years (5, 10, 15, and 20) respectively. Several vital factors such as BMI, LDL-level, HDL-level, blood pressure, exercise level, etc. are measured along with family history, medical history, physical activity, nutrient intake, obesity questions, pyschosocial, pulmonary function etc.. We use Statistical Relational Learning (SRL) algorithms in order to automatically estimate models for predicting the Coronary Artery Calcification (CAC) levels, a measure of subclinical CAD, in year 20 (corresponding to 2005 when the subjects were between 38 and 50 years old) given the measurements from all the previous years. Using the predictions of the CAC levels, we can predict cardiovascular events such as heart attacks. In turn, this allows us to enable pro-active treatment planning for the high-risk patients. That is, to identify young adult patients who are at a potentially high risk for cardiovascular events and design patient-specific treatments that will mitigate the risks.

• Members involved:
• Sriraam Natarajan and Santiago Saldana
• Other collaborators:
• Dr. Jeff Carr (Wake Forest School of Medicine) and Dr. Edward Ip (Wake Forest School of Medicine)

• Publications to be updated soon.

Assistive Technology for the Disabled

We are currently working on building a domain-independent platform that will serve as the basis for several possible collaborations. This platform will enable end-to-end research in mobile applications as well as involve several steps ranging from data collection using sensor inputs to the building and analysis of predictive models. Such an infrastructure will enable us to work on different problems such as building instructional interfaces for children with special needs, providing assistance to people with disabilities, providing appropriate prompts to people with dementia, etc.. For example, the Verbal Victor iPhone application developed by Dr. Paul Pauca of the Wake Forest University Computer Science Department presents a set of images to a child and the child can choose the appropriate image for communicating with others. In particular, the application automatically presents the recorded sound corresponding to the chosen image. This is currently a one-way communication. We are planning on extending this to take the feedback from the children (which is implicit) in order to present a personalized set of images for each child based on several features such as the time of day (indicating whether the child is requesting lunch or play), the activities performed until a given instance of time (for instance, the child might require a nap after eating), etc.. To do so, we need to integrate the sensory input (in this case, the child's selection) with the analysis of the chosen images and predict the set of appropriate images to present to the child. Our research group is in the process of prototyping such an assistive communication tool using conventional mobile technology such as the iPad instead of dedicated devices. To our knowledge, this tool is the first of its kind and with your support, it could be released as early as next year. Given that the current version of Verbal Victor already has more than 1500 downloads at the cost of $7 each, we anticipate that a personalizable Smart Verbal Victor would have an even bigger impact in the community.

• Members involved:
• Sriraam Natarajan, Phillip Odom
• Other collaborators:
• Dr. Paul Pauca (Wake Forest University)

• Publications to be updated soon.

Bio-Toxicity in Cancer Patients

The group is also involved in identifying the sub-group of cancer patients that are susceptible to cardiovascular risks. We are currently employing SRL algorithms on a set of 1000 patients who are undergoing cancer treatment. Some of these patients have exhibited cardiovascular issues. Our goal is to use the demographic information and prescription information of these patients in order to isolate the subset of patients who are at risk.

• Members involved:
• Sriraam Natarajan, Adam Edwards
• Other collaborators:
• Dr. Yaorong Ge (Wake Forest School of Medicine) and Dr. Greg Hundley (Wake Forest School of Medicine)

• Publications to be updated soon.

Identifying Patient Status from MRI Images

The group is also involved in using machine learning algorithms to identify the status of patients using 3-D MRI images. Magnetic resonance imaging (MRI) has emerged as an important tool to identify intermediate biomarkers of Alzheimer's disease (AD) due to its ability to measure regional changes in the brain that are thought to reflect disease severity and progression. In this paper, we set out a novel data mining pipeline that uses volumetric MRI data collected from different subjects as input and classifies them into one of three classes: AD, mild cognitive impairment (MCI) and cognitively normal (CN). Our pipeline consists of three stages - (1) a segmentation layer where brain MRI data is divided into clinically relevant regions; (2) a classification layer that uses relational learning algorithms to make pairwise predictions between the three classes; and (3) a combination layer that combines the results of the different classes to obtain the final classification. One of the key features of our proposed approach is that it allows for domain expert's knowledge to guide the learning in all the layers. We evaluate our pipeline on 397 patients acquired from the Alzheimer's Disease Neuroimaging Initiative and demonstrate that it obtains state-of-the-art performance with minimal feature engineering.

• Members involved:
• Sriraam Natarajan, Saket Joshi, Baidya Saha, Adam Edwards, and Tushar Khot
• Other collaborators:
• Elizabeth Moody, Kristian Kersting, Dr. Joseph Maldjian (Wake Forest School of Medicine) and Dr. Chris Whitlow (Wake Forest School of Medicine)

• Publications to be updated soon.