Session 2: Processing architectures for biomedical signal and image processing

Abstract

Biomedical Signal and Image Processing presents a variety of challenges in terms of the dimensionality of the data being processed, nature of the processing involved, the expected data throughput and the peculiar requirements for embedded applications. The processing involved often entails the detection of patterns of interest such as lesions in tissue and abnormal transients or rhythms in ECG or EEG data. The required data throughput can be very large in the case of multi-channel or multi-dimensional time-varying signals. While General Purpose Processor and Graphical Processing Unit implementations abound, custom architectures can be necessary to accelerate processing in order for time-critical information to be available to a clinician. For embedded biomedical applications, energy efficiency severely constrains custom processor design. For implantable devices, these constraints are even more stringent. This special session includes two excellent papers that address some of these issues. Both consider the implementation of biomedical signal processing algorithms with a view toward an eventual implantable device. One paper focuses on neural signal decoders to control artificial limbs, and the other is concerned with the analysis of ECG signals, in particular the detection of QRS complexes as a pre-processing step to detect various abnormal heart conditions.