Prediction of foot clearance parameters as a precursor to forecasting the risk of tripping and falling

Tripping and falling is a serious health problem for older citizens due to the high medical costs incurred and the high mortality rates precipitated mostly by hip fractures that do not heal well. Current falls prevention technology encompasses a broad range of interventions; both passive (e.g., safer environments, hip protectors) and active (e.g., sensor-based fall detectors) which attempt to reduce the effects of tripping and falling. However the majority of these interventions minimizes the impact of falls and do not directly reduce the risk of falling. This paper investigates the prediction of gait parameters related to foot-to-ground clearance height during the leg swing phase which have been physically associated with tripping and falling risk in the elderly. The objective is to predict parameters of foot trajectory several walking cycles in advance so that anticipated low foot clearance could be addressed early with more volitional countermeasures, e.g., slowing down or stopping. In this primer study, foot kinematics was recorded with a highly accurate motion capture system for 10 healthy adults (25-32 years) and 11 older adults (65-82 years) with a history of falls who each performed treadmill walking for at least 10 min. Vertical foot displacement during the swing phase has three characteristic inflection points and we used these peak values and their normalized time as the target prediction values. These target variables were paired with features extracted from the corresponding foot acceleration signal (obtained through double differentiation). A generalized regression neural network (GRNN) was used to independently predict the gait variables over a prediction horizon (number of gait cycles ahead) of 1-10 gait cycles. It was found that the GRNN attained 0.32-1.10 cm prediction errors in the peak variables and 2-8% errors in the prediction of normalized peak times, with slightly better accuracies in the healthy group compared to elderly fallers. Prediction accuracy decreased linearly (best fit) at a slow rate with increasing prediction horizon ranging from 0.03 to 0.11 cm per step for peak displacement variables and 0.34 × 10(-3) - 1.81 × 10(-3)% per step for normalized peak time variables. Further time series analysis of the target gait variable revealed high autocorrelations in the faller group indicating the presence of cyclic patterns in elderly walking strategies compared to almost random walking patterns in the healthy group. The results are promising because the technique can be extended to portable sensor-based devices which measure foot accelerations to predict the onset of risky foot clearance, thus leading to a more effective falls prevention technology.

On the feasibility of learning to predict minimum toe clearance under different walking speeds

A major concern in human movement research is preventing tripping and falling which is known to cause severe injuries and high fatalities in elderly (>65 years) populations. Current falls prevention technology consists of active interventions e.g., strength and balance exercises, preimpact fall detectors, and passive interventions e.g., shower rails, hip protectors. However it has been found that these interventions with the exception of balance exercises do not effectively reduce falls risk. Recent work has shown that the minimum toe clearance (MTC) can be successfully monitored to detect gait patterns indicative of tripping and falling risk. In this paper, we investigate the feasibility of predicting MTC values of consecutive gait cycles under different walking speeds. The objective is two-fold, first to determine if end point foot trajectories can be accurately predicted and second, if walking speed is a significant parameter which influences the prediction process. The Generalized Regression Neural Networks and the Support Vector Regressor models were trained to predict MTC time series successively over an increasing prediction horizon i.e., 1 to 10 steps. Increased walking speeds resulted in increased MTC variability but no significant increase in mean MTC height. Root mean squared prediction errors ranged between 2.2-2.6mm or 10% of the mean values of the respective test data. The SVM slightly outperformed the GRNN predictions (0.5%-2.1% better accuracy). Best prediction accuracies decreased by 0.5mm for a doubling of walking speed i.e., from 2.5 km/h to 5.5 km/h. The results are encouraging because they demonstrate that the technique could be applied to forecasting low MTC values and provide new approaches to falls prevention technologies.

Artificial neural network modeling of apoptosis in gammairradiated human lymphocytes

PURPOSE

To develop an artificial neural network (ANN) model of apoptotic response in gamma irradiated human lymphocytes. To assess the feasibility of training ANN radiobiological models using data collected with flow cytometry.

MATERIALS AND METHODS

Irradiated isolated human lymphocytes were labelled with Annexin V-Fluorescein Isothiocyanate (FITC) and 7-Amino-Actinomycin D (7AAD) then analysed using flow cytometry. Twenty-four dose responses per donor from 14 donors were collected from a flow cytometer and used in model development as the training and cross-validation datasets. The general ANN model architecture was a multi-layer perceptron using the mean squared error of a cross validation dataset as the objective function. The ANN model was optimized by varying the number of hidden layers and the number of processing elements per layer. The optimized model constituted of three hidden layers with 80, 40, and 10 hidden layers in the first, second, and third layers respectively.

RESULTS

The optimized model was used to simulate dose responses at the training doses of 0, 2, 4 and 8 Gray. A strong agreement between the model and measured dose responses was observed. The model was also used to simulate a dose response at 0.1 Gray and results were compared to the measured dose response from a donor not used in model development. Again, strong agreement between the model and the observed dose response was found.

CONCLUSIONS

This study shows that artificial neural networks can be trained to provide high resolution, high accuracy models of multivariate radiobiological data collected by flow cytometry.

A Hybrid Neurogenetic Approach for Stock Forecasting

In this paper, we propose a hybrid neurogenetic system for stock trading. A recurrent neural network (NN) having one hidden layer is used for the prediction model. The input features are generated from a number of technical indicators being used by financial experts. The genetic algorithm (GA) optimizes the NN's weights under a 2-D encoding and crossover. We devised a context-based ensemble method of NNs which dynamically changes on the basis of the test day's context. To reduce the time in processing mass data, we parallelized the GA on a Linux cluster system using message passing interface. We tested the proposed method with 36 companies in NYSE and NASDAQ for 13 years from 1992 to 2004. The neurogenetic hybrid showed notable improvement on the average over the buy-and-hold strategy and the context-based ensemble further improved the results. We also observed that some companies were more predictable than others, which implies that the proposed neurogenetic hybrid can be used for financial portfolio construction.

Stock trading using RSPOP: a novel rough set-based neuro-fuzzy approach

This paper investigates the method of forecasting stock price difference on artificially generated price series data using neuro-fuzzy systems and neural networks. As trading profits is more important to an investor than statistical performance, this paper proposes a novel rough set-based neuro-fuzzy stock trading decision model called stock trading using rough set-based pseudo outer-product (RSPOP) which synergizes the price difference forecast method with a forecast bottleneck free trading decision model. The proposed stock trading with forecast model uses the pseudo outer-product based fuzzy neural network using the compositional rule of inference [POPFNN-CRI(S)] with fuzzy rules identified using the RSPOP algorithm as the underlying predictor model and simple moving average trading rules in the stock trading decision model. Experimental results using the proposed stock trading with RSPOP forecast model on real world stock market data are presented. Trading profits in terms of portfolio end values obtained are benchmarked against stock trading with dynamic evolving neural-fuzzy inference system (DENFIS) forecast model, the stock trading without forecast model and the stock trading with ideal forecast model. Experimental results showed that the proposed model identified rules with greater interpretability and yielded significantly higher profits than the stock trading with DENFIS forecast model and the stock trading without forecast model.

Time dependent neural network models for detecting changes of state in complex processes: Applications in earth sciences and astronomy

A computational intelligence approach is used to explore the problem of detecting internal state changes in time dependent processes; described by heterogeneous, multivariate time series with imprecise data and missing values. Such processes are approximated by collections of time dependent non-linear autoregressive models represented by a special kind of neuro-fuzzy neural network. Grid and high throughput computing model mining procedures based on neuro-fuzzy networks and genetic algorithms, generate: (i) collections of models composed of sets of time lag terms from the time series, and (ii) prediction functions represented by neuro-fuzzy networks. The composition of the models and their prediction capabilities, allows the identification of changes in the internal structure of the process. These changes are associated with the alternation of steady and transient states, zones with abnormal behavior, instability, and other situations. This approach is general, and its sensitivity for detecting subtle changes of state is revealed by simulation experiments. Its potential in the study of complex processes in earth sciences and astrophysics is illustrated with applications using paleoclimate and solar data.

Bi-directional computing architecture for time series prediction

A number of neural network models and training procedures for time series prediction have been proposed in the technical literature. These models studied for different time-variant data sets have typically used uni-directional computation flow or its modifications. In this study, on the contrary, the concept of bi-directional computational style is proposed and applied to prediction tasks. A bi-directional neural network model consists of two subnetworks performing two types of signal transformations bi-directionally. The networks also receive complementary signals from each other through mutual connections. The model not only deals with the conventional future prediction task, but also with the past prediction, an additional task from the viewpoint of the conventional approach. An improvement of the performance is achieved through making use of the future-past information integration. Since the coupling effects help the proposed model improve its performance, it is found that the prediction score is better than with the traditional uni-directional method. The bi-directional predicting architecture has been found to perform better than the conventional one when tested with standard benchmark sunspots data.

Recurrent neural network-based approach for early recognition of Alzheimer's disease in EEG

OBJECTIVE

We explored the ability of specifically designed and trained recurrent neural networks (RNNs), combined with wavelet preprocessing, to discriminate between the electroencephalograms (EEGs) of patients with mild Alzheimer's disease (AD) and their age-matched control subjects.

METHODS

Twomin recordings of resting eyes-closed continuous EEGs (as well as their wavelet-filtered subbands) obtained from parieto-occipital channels of 10 early AD patients and 10 healthy controls were input into RNNs for training and testing purposes. The RNNs were chosen because they can implement extremely non-linear decision boundaries and possess memory of the state, which is crucial for the considered task.

RESULTS

The best training/testing results were achieved using a 3-layer RNN on left parietal channel level 4 high-pass wavelet subbands. When trained on 3 AD and 3 control recordings, the resulting RNN tested well on all remaining controls and 5 out of 7 AD patients. This represented a significantly better than chance performance of about 80% sensitivity at 100% specificity.

CONCLUSION

The suggested combined wavelet/RNN approach may be useful in analyzing long-term continuous EEGs for early recognition of AD. This approach should be extended on larger patient populations before its clinical diagnostic value can be established. Further lines of investigation might also require that EEGs be recorded from patients engaged in certain mental (cognitive) activities.