Compressive Sensing Applications to Biomedical Engineering

点击翻译按钮获取中文摘要

Compressed Sensing (CS) has been successfully applied to MRI. While MRI remains an intersting venue for CS research, other biomedical applications could also benefit from CS theory. This proposal explores the application of CS to new research problems in MRI, X-Ray Computed Tomography and energy efficient EEG signal transmission. These problems are novel but all fall under the broad category of CS. The outcomes will benefit the biomedical engineering community as well as strengthen research in CS as a whole. Dynamic MRI Reconstruction in real-time involves “fast“ processing of many frames per second. Existing reconstruction techniques are offline and only useful for analytical tasks which can be done posthumously, like medical diagnosis and neurological studies. There are other major applications which require real-time reconstruction – image-guided surgery and tracking / monitoring applications. Real-time dynamic MRI reconstruction is a hard problem and has received limited focus so far. There is a need to address this problem and develop efficient, robust and accurate techniques for real-time reconstruction. Reducing Ionizing Radiation for dynamic X-Ray CT results in 30,000 cases of cancer a year and about 15,000 deaths from cancer in the USA. CT is not safe .The problem is aggravated in dynamic CT, as the patient is subjected to more ionizing radiation compared to static scans. We will develop new techniques to address the dire need to reduce the ionizing radiation in CT. Energy efficient EEG transmission for Wireless Body Area Network (WBAN) : In Canada 14.1% of the population is above the age of 65. They should be able to live with dignity; with minimum dependency on others. At the same time they need to be monitored for health conditions. From EEG signals it is possible to infer a variety of health problems. We envision a system where the EEG signal will be acquired at the subject’s location and then transmitted to a healthcare unit for monitoring and analysis. Since the battery life is limited in WBAN applications we will design sampling and transmission protocols that are energy efficient. The MRI, X-Ray CT and EEG problems are tied by a common goal – how to reconstruct the underlying signal from a reduced number of measurements. Thus each problem will be first recast in the CS framework and methodolgies for its solutions will be developed. Real-time dynamic MRI reconstruction is presently solved via two approaches. The first uses dynamical system models like Kalman Filtering, which is not computationally or memory efficient. The other uses CS. This yields more accurate results but remains too slow for real-time performance. We propose to combine the two in a prediction-correction framework. In the prediction step a dynamical model will be used to estimate the frames; in the correction step the predicted estimate will be refined using CS. Offline dynamic MRI reconstruction is well studied but for dynamic CT, only a handful of studies exist. We will leverage our expertise in dynamic MRI techniques to suit the needs for CT. We plan to model the dynamic CT frames as a Casorati matrix. This matrix is sparse in transform domain and will also be low-rank. The novel Casorati matrix model will enable us to exploit both transform domain sparsity as well as its low-rank structure to obtain results with lesser ionizing radiation. The problem of EEG transmission over WBAN is not a mature topic. Signal processing researchers used CS to reduce the number of samples to be transmitted– for power communication Communication theory experts tried encoding techniques for efficient transmission whereas researchers in sensor networks used novel switching techniques to save energy. We want to address the problem as a whole (end-to-end) .