The current applications of neural networks to in vivo medical imaging and signal processing are reviewed. As is evident from the literature neural networks have already been used for a wide variety of tasks within medicine. As this trend is expected to continue this review contains a description of recent studies to provide an appreciation of the problems associated with implementing neural networks for medical imaging and signal processing.

Review of neural network applications in medical imaging and signal processing.

Applications of artificial neural nets in clinical biomechanics.

Describes a binary adaptive resonance theory (ART-1)-based artificial neural network (ANN) adapted for controlling functional electrical stimulation (FES) to facilitate patient-responsive ambulation by paralyzed patients with spinal cord injures. This network is to serve as a controller in an FES system developed by the first author which is presently in use by 300 patients worldwide (still without ANN control) and which was the first and the only FES system approved by the FDA. The proposed neural network discriminates above-lesion upper-trunk electromyographic (EMG) time series to activate standing and walking functions under FES and controls FES stimuli levels using response-EMG signals. For this particular application, the authors introduce several modifications of the ART-1 for pattern recognition and classification. First, a modified on-line learning rule is proposed. The new rule assures bidirectorial modification of the stored patterns and prevents noise interference. Second, a new reset rule is proposed which prevents "exact matching" when the input is a subset of the chosen pattern. The authors show the applicability of a single ART-1-based structure to solving two problems, namely, 1) signal pattern recognition and classification, and 2) control. This also facilitates ambulation of paraplegics under FES, with adequate patient interaction in initial system training, retraining the network when needed, and in allowing patient's manual override in the ease of error, where any manual override serves as a retraining input to the neural network. Thus, the practical control problems (arising in actual independent patient ambulation via FES) were all satisfied by a relatively simple ANN design. >

Artificial neural network control of FES in paraplegics for patient responsive ambulation

https://estudogeral.sib.uc.pt/bitstream/10316/9993/1/ficheiro.pdf

Brief paper: Hand movement recognition based on biosignal analysis

The paper describes an ART-1-based neural network adapted for controlling functional electrical stimulation (FES) to facilitate patient-responsive ambulation by paralyzed patients with spinal cord injuries. The proposed neural network discriminates above-lesion upper-trunk EMG time series to activate standing and walking functions under FES and controls FES stimuli levels using response-EMG signals. For this particular application we introduce several modifications of ART-1 theory for pattern recognition and classification. We show the applicability of a single ART-1-based structure to solving two problems: (a) signal pattern recognition and classification and (b) control. This also facilitates ambulation of paraplegics under FES, with adequate patient interaction in initial system training, retraining the network when needed, and in allowing patient's manual over-ride in the case of error, where any manual over-ride serves as a re-training input to the neural network. >

Neural network control of FES in paraplegics for patient-responsive ambulation

In this paper, we discuss the application of a back propagation neural network to the classification of electromyogram (EMG) signal patterns for the control of prosthesis. In our approach, the collection process of EMG information from remaining muscles of the amputee is performed through a amputee fuzzy training procedure using the information given by human supervisor. Multilayer back propagation model is implemented to a two degrees of freedom above elbow prosthesis which shows a high classification performance. The model is expected to be extended well to the cases of higher degrees if the systematic study will be done on the fuzzy contraction states of muscles, feature selection process of EMG signals and appropriate structure of the proposed network.

EMG Pattern Classification Based On Back Propagation Neural Network For Prosthesis Control

In this paper, a heuristic approach for classification of the electromyogram (EMG) signal patterns for the control of a two degrees of freedom above elbow arm prosthesis is presented. The rule set has been deduced from a close observation of the trajectory of EMG pattern vector. The basic idea is to take into consideration both present and past position of the pattern vector. At each decision period, a score is given to each motion class using an appropriate rule set and then the class which holds the highest score is selected. EMG signals were picked up from the remaining muscles of the stump such as the biceps and triceps. The selected feature was the integral absolute value (IAV) of the EMG. Experimental results on amputee which are presented show a very low misclassification rate.

http://ieeexplore.ieee.org/iel4/499/4080/00692243.pdf

Electromyogram Pattern Classification Based On Heuristic Rules

Although neural nets have shown their ability for the EMG pattern classification in order to control the cybernetic arm, however, the s election of topology as well as parameters of the network is still a matter of question. This paper is a step towards solving this problem through simulation of a numer of networks. Priliminary results show that back-propagation network results into a better performance.

http://ieeexplore.ieee.org/iel4/1015/7498/00684546.pdf

A Comparative Study Of Neural Networks Models For EMG Pattern Classification

In order to design an EMG-controlled cybernetic arm with high degree of freedom, the optimal subset of the possible muscle contraction states is to be determined for the purpose of controlling the arm. In this paper, optimal muscle contraction states have been selected using the classification error in probabilistic neural network (PNN). Results of simulating the network show the fact that reliable contraction states can be obtained for representing movement classes for both cases of two and three degree of freedom arm.

http://ieeexplore.ieee.org/iel4/1015/7498/00684044.pdf

Optimal Selection Of Muscle Contraction States Using Probabilistic Neural Myoprocessor

A combined connectionist and symbolic approach consisting of a rule-base and a neural network has been presented for interpreting motion patterns in a cybernetic arm. In order to realize the utility domain of each approach, a situation vector is identified at the beginning of interpretatio n process. Results demonstrate the fact that interpretation performance can be increased for complex motions if the two approaches are utilized in their proper utility domain.

http://ieeexplore.ieee.org/iel4/499/4080/00691843.pdf

T. Khoshaba

Papers

EMG Pattern Classification Based On Back Propagation Neural Network For Prosthesis Control

Electromyogram Pattern Classification Based On Heuristic Rules

A Comparative Study Of Neural Networks Models For EMG Pattern Classification

Optimal Selection Of Muscle Contraction States Using Probabilistic Neural Myoprocessor

A Combined Connectionist And Symbolic Approach To Motion Interpretation In Cybernetic Arm