Omid Dehzangi

By | | No Comments

Wearable health technology is drawing significant attention for good reasons. The pervasive nature of such systems providing ubiquitous access to the continuous personalized data will transform the way people interact with each other and their environment. The resulting information extracted from these systems will enable emerging applications in healthcare, wellness, emergency response, fitness monitoring, elderly care support, long-term preventive chronic care, assistive care, smart environments, sports, gaming, and entertainment which create many new research opportunities and transform researches from various disciplines into data science which is the methodological terminology for data collection, data management, data analysis, and data visualization. Despite the ground-breaking potentials, there are a number of interesting challenges in order to design and develop wearable medical embedded systems. Due to limited available resources in wearable processing architectures, power-efficiency is demanded to allow unobtrusive and long-term operation of the hardware. Also, the data-intensive nature of continuous health monitoring requires efficient signal processing and data analytic algorithms for real-time, scalable, reliable, accurate, and secure extraction of relevant information from an overwhelmingly large amount of data. Therefore, extensive research in their design, development, and assessment is necessary. Embedded Processing Platform Design The majority of my work concentrates on designing wearable embedded processing platforms in order to shift the conventional paradigms from hospital-centric healthcare with episodic and reactive focus on diseases to patient-centric and home-based healthcare as an alternative segment which demands outstanding specialized design in terms of hardware design, software development, signal processing and uncertainty reduction, data analysis, predictive modeling and information extraction. The objective is to reduce the costs and improve the effectiveness of healthcare by proactive early monitoring, diagnosis, and treatment of diseases (i.e. preventive) as shown in Figure 1.

dehzangi-image

Embedded processing platform in healthcare

Issam El Naqa

By | | No Comments

Our lab’s research interests are in the areas of oncology bioinformatics, multimodality image analysis, and treatment outcome modeling. We operate at the interface of physics, biology, and engineering with the primary motivation to design and develop novel approaches to unravel cancer patients’ response to chemoradiotherapy treatment by integrating physical, biological, and imaging information into advanced mathematical models using combined top-bottom and bottom-top approaches that apply techniques of machine learning and complex systems analysis to first principles and evaluating their performance in clinical and preclinical data. These models could be then used to personalize cancer patients’ chemoradiotherapy treatment based on predicted benefit/risk and help understand the underlying biological response to disease. These research interests are divided into the following themes:

  • Bioinformatics: design and develop large-scale datamining methods and software tools to identify robust biomarkers (-omics) of chemoradiotherapy treatment outcomes from clinical and preclinical data.
  • Multimodality image-guided targeting and adaptive radiotherapy: design and develop hardware tools and software algorithms for multimodality image analysis and understanding, feature extraction for outcome prediction (radiomics), real-time treatment optimization and targeting.
  • Radiobiology: design and develop predictive models of tumor and normal tissue response to radiotherapy. Investigate the application of these methods to develop therapeutic interventions for protection of normal tissue toxicities.
Machine Learning in Radiation Oncology: Theory and Applications

Machine Learning in Radiation Oncology: Theory and Applications

Laura Balzano

By | | No Comments

Professor Balzano and her students investigate problems in statistical signal processing and optimization, particularly dealing with large and messy data. Her applications typically have missing, corrupted, and uncalibrated data as well as heterogeneous data in terms of sensors, sensor quality, and scale in both time and space. Her theoretical interests involve classes of non-convex problems that include Principal Components Analysis (or the Singular Value Decomposition) and many interesting variants such as PCA with sparse or structured principal components, orthogonality and non-negativity constraints, and even categorical data or human preference data. She concentrates on fast gradient methods and related optimization methods that are scalable to real-time operation and massive data. Her work provides algorithmic and statistical guarantees for these algorithms on the aforementioned non-convex problems, and she focuses carefully on assumptions that are realistic for the relevant applications. She has worked in the areas of online algorithms, real-time computer vision, compressed sensing and matrix completion, network inference, and sensor networks.

Real-time dynamic background tracking and foreground separation. At time t = 101, the virtual camera slightly pans to right 20 pixels. We show how GRASTA quickly adapts to the new subspace by t = 125. The first row is the original video frame; the middle row is the tracked background; the bottom row is the separated foreground.

Real-time dynamic background tracking and foreground separation. At time t = 101, the virtual camera slightly pans to right 20 pixels. We show how GRASTA quickly adapts to the new subspace by t = 125. The first row is the original video frame; the middle row is the tracked background; the bottom row is the separated foreground.

Alfred Hero

By | | No Comments

Alfred O. Hero, PhD, is the R. Jamison and Betty Williams Professor of Engineering at the University of Michigan and co-Director of the Michigan Institute for Data Science.

The Hero group focuses on building foundational theory and methodology for data science and engineering. Data science is the methodological underpinning for data collection, data management, data analysis, and data visualization. Lying at the intersection of mathematics, statistics, computer science, information science, and engineering, data science has a wide range of application in areas including: public health and personalized medicine, brain sciences, environmental and earth sciences, astronomy, materials science, genomics and proteomics, computational social science, business analytics, computational finance, information forensics, and national defense. The Hero group is developing theory and algorithms for data collection, analysis and visualization that use statistical machine learning and distributed optimization. These are being to applied to network data analysis, personalized health, multi-modality information fusion, data-driven physical simulation, materials science, dynamic social media, and database indexing and retrieval. Several thrusts are being pursued:

  1. Development of tools to extract useful information from high dimensional datasets with many variables and few samples (large p small n). A major focus here is on the mathematics of “big data” that can establish fundamental limits; aiding data analysts to “right size” their sample for reliable extraction of information. Areas of interest include: correlation mining in high dimension, i.e., inference of correlations between the behaviors of multiple agents from limited statistical samples, and dimensionality reduction, i.e., finding low dimensional projections of the data that preserve information in the data that is relevant to the analyst.
  2. Data representation, analysis and fusion on non-linear non-euclidean structures. Examples of such data include: data that comes in the form of a probability distribution or histogram (lies on a hypersphere with the Hellinger metric); data that are defined on graphs or networks (combinatorial non-commutative structures); data on spheres with point symmetry group structure, e.g., quaternion representations of orientation or pose.
  3. Resource constrained information-driven adaptive data collection. We are interested in sequential data collection strategies that utilize feedback to successively select among a number of available data sources in such a way to minimize energy, maximize information gains, or minimize delay to decision. A principal objective has been to develop good proxies for the reward or risk associated with collecting data for a particular task (detection, estimation, classification, tracking). We are developing strategies for model-free empirical estimation of surrogate measures including Fisher information, R'{e}nyi entropy, mutual information, and Kullback-Liebler divergence. In addition we are quantifying the loss of plan-ahead sensing performance due to use of such proxies.
Correlation mining pipeline transforms raw high dimensional data (bottom) to information that can be rendered in interpretable sparse graphs and networks, simple screeplots, and denoised images (top). The pipeline controls data collection, feature extraction and correlation mining by integrating domain information and its assessed value relative to the desired task (on left) and accounting for constraints on data collection budget and uncertainty bounds (on right).

Correlation mining pipeline transforms raw high dimensional data (bottom) to information that can be rendered in interpretable sparse graphs and networks, simple screeplots, and denoised images (top). The pipeline controls data collection, feature extraction and correlation mining by integrating domain information and its assessed value relative to the desired task (on left) and accounting for constraints on data collection budget and uncertainty bounds (on right).