Statistical Signal Processing

John R. Deller  | |


Research in the Statistical Signal Processing (SSP) Laboratory covers theory and algorithms of statistical signal processing; detection, estimation, parametric modeling and system identification; nonlinear modeling of functional and effective connectivity in the brain; sensor fusion for detection and assessment of Alzheimer's disease; evolutionary monitoring of schizophrenic spectrum disorders; speech analysis, recognition, and coding- theory and applications.

In this new “Century of Biology,” classical linear, time-invariant models will not support research into the complex systems that characterize life.    Nowhere is this more true than in the quest to unravel the mysteries of the human brain.  The SSP Lab has turned its longstanding interest in adaptive nonlinear modeling to research into the methods by which the human brain stores, codes, and communicates information.  This work, which is in collaboration with Profs. Selin Aviyente and Kalyan Deb of ECE, and Erik Goodman at MSU's BEACON Center, features the following:

  • A novel approach to nonlinear system identification integrating four three modeling and identification strategies: linear-time-invariant-in-parameters models, set-based parameter identification, evolutionary algorithms, and multi-objective optimization over fitness measures derived from the set solutions.
  • Nonlinear modeling for effective brain network connectivity, based on Granger causality in complex models.
  • Significant potential for future theoretical development and wide applicability to nonlinear filtering, detection, and estimation, and to pathological state feature extraction and identification, for an array of biomedical modeling problems.

Statistical Image Analysis for Detection and Prognosis of Alzheimer's Disease

SSP Researchers in collaboration with colleagues at Beijing University of Technology, the BEACON Center at MSU, and MSU's College of Human Medicine,  are developing effective methods for the pixel-level fusion of images created from multiple sensor technologies. This image-processing research is based on recent developments in the mathematical theories of matrix completion and sparse data representations.   SSP Lab researchers are exploring ways to extend these image-fusion methods for deployment in a very different set of problems in biomedicine.  The initial aim of the new work is to develop “fusion” methods for a variety multi-sensor diagnostic imaging and other information sources in the study of mental health conditions like autism, depression, and Alzheimer's disease. 

  • One of the central objectives is to incorporate evolutionary computational methods into the fusion strategies in an effort to discover new and subtle connections that may exist in the disparate medical data which reflect the complex human nervous system. 
  • A specific application seeks to model longitudinal MRI imaging biomarkers in diseased brains in order to project the time course of the disease before symptoms are experienced by the patient.  
  • The data fusion and matrix completion theories are being researched for the potential to extrapolate, but also interpolate, individual patient medical records (e.g., missing scans) from information inferred from a large patient population of disease histories.