Proceedings of the 19th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. 'Magnificent Milestones and Emerging Opportunities in Medical Engineering' (Cat. No.97CH36136)

Head movement has been used as a control interface for people with motor impairments in a range of applications. Chin operated joysticks and switch arrays have been incorporated in control systems for electric wheelchairs but have several disadvantages, including being difficult to operate and aesthetically unattractive. A prototype wheelchair control interface has been developed that makes use of an artificial neural network (ANN) to recognize commands given by head movement. This paper presents the results of an experimental investigation of the ANN's performance in terms of classification accuracy and delay. It goes on to compare the results of disabled with able-bodied users, and assesses the effect of providing real-time feedback to the user. The results obtained indicate that ANN techniques can be used to classify head movements sufficiently quickly and accurately to be used in a practical interface. The provision of graphical real-time feedback does not appear to be crucial, but may be of benefit for particular cases.

The Poincare plot is an emerging Heart Rate Variability (HRV) analysis technique, the geometry of which has been shown to distinguish between healthy and unhealthy subjects in clinical settings. The Poincare plot is able to display nonlinear aspects of the interval sequence and is therefore of interest in characterizing the nonlinear aspects of HRV. The problem is, how do we quantitatively characterize the geometry of the plot to capture useful descriptors that are independent of existing HRV measures? In this paper, we investigate a popular existing category of techniques and show that they measure linear aspects of the intervals which existing HRV indices already specify. The fact that these methods appear insensitive to the nonlinear characteristics of the intervals is an important finding because the Poincare plot is primarily a nonlinear technique.

Predicting the onset of hypoglycaemia can avoid major health complications in Type I insulin-dependent-diabetes-mellitus (IDDM) patients. This paper describes the design of a novel fuzzy neural network estimator algorithm (FNNE) for predicting the glycaemia profile and onset of hypoglycaemia in insulin-induced subjects, by modelling the changes in heart rate and skin impedance parameters. Hypoglycaemia was induced briefly in 12 volunteers (group A: 6 non-diabetic subjects and group B: 6 Type 1 IDDM patients) using insulin infusion. Their skin impedances, heart rates and actual blood glucose levels (BGL) were monitored at regular intervals. The FNNE algorithm was trained using all subjects from group A and validated/tested on the remaining subjects from group B. The mean error of estimation of BGL profile for the training data set (group A) was 0.107 (p < 0.05) and for the validation/test data set (group B) was 0.139 (p < 0.05). Furthermore, the FNNE algorithm was able to predict the onset of hypoglycaemia episodes in group A and group B with a mean error of 0.071 (p < 0.03) and 0. 176 (p < 0.05) respectively.

Hypoglycaemia is the most common complication experienced by patients with Type 1 insulin dependent diabetes mellitus (IDDM). During hypoglycaemia, physiological parameter changes occur in patients, due to the sympathetic nervous system, of which the most predominant parameters are sweating and heart rate. Non-invasive monitoring and detecting hypoglycaemia in diabetic patients requires simultaneous measurements of body functions such as sweating, breathing patterns, recording EEG and heart rate. A hypoglycaemic monitor has been developed to help the physician predict the onset of hypoglycaemia. This paper describes the development of a fuzzy estimator as the main software engine of the hypoglycaemic monitor. The fuzzy estimator consists of an initial static estimator and has been expanded using self organisation to enhance it's flexibility.

Evaluation of biomedical signals is important in the diagnosis of numerous diseases, chiefly in cardiology through the use of electrocardiograms, and to a more limited extent, in neurology through the use of electroencephalograms. While automated techniques exist for both ECG and EEG analysis, it is likely that additional information can be extracted from these signals through the use of new methods. A chaotic method for analysis of signal analysis variability is presented here that identifies the degree of variability in the signal over time. A second focus is to develop higher order decision models that can incorporate these results with other clinical parameters to represent a more comprehensive view of the disease state, using a neural network model.

The performance of the three-shell spherical head model versus the performance of the realistic head model is investigated when solving the inverse problem with a single dipole, in the presence of noise. This is evaluated by inspecting the average dipole location error when applying a spherical and a realistic head model, for 1000 noisy scalp potentials, originating from the same test dipole and having the same noise level. The location errors are obtained utilizing a local linearization, which is validated with a Monte Carlo simulation. For 27 electrodes, an EEG epoch of one time sample and spatially white Gaussian noise we found that the importance of the realistic head model over the spherical head model reduces by increasing the noise level.

We have explored the feasibility of an alternative strategy using biological signals such as sEMG of the sternocleidomastoid muscle (SCM) for variable control of our head movement system. Seven volunteers were instrumented with bilateral sEMG sensors on the SCM. Basic neck movements of lateral tilts and graded head rotations were performed. Data were normalized as a percentage of maximum voluntary contractions (MVC) for right and left sides, respectively. The contribution from ipsilateral sEMG signal as percentage of full-range was /spl sim/75% for left and 55% for right head tilts. During head rotations at 30, 45, and 60/spl deg/ to both sides, results for sEMG signal amplitude as a percentage of MVC showed excellent reproducibility of the contralateral SCM at approximately 10%, 18%, and 32% on both sides. Despite the small number of subjects for a thorough statistical analysis, no differences exist in t-tests between sEMG (as % of MVC) right and left sides during head rotation; however, differences do exist for each level of rotation (p<0.01). Head rotation provided the most consistent sEMG signal correlation with the degree of motion in all subjects, allowing for reproducible proportional control for our head movement technology.

Signal quantitation in most near infrared-spectroscopy (NIRS) instruments is achieved through solving simultaneous equations or multiple regression analysis. The aim of this study was to compare NIRS signal quantitation by conventional multiple regression to artificial neural networks. Sixteen adult sheep were used in the study of the effects of changes in cerebral blood flow and metabolism through induction of seizures, ischemia, and hypercapnia. NIRS-derived signal attenuation for relative blood volume (BV) and oxygen desaturation (DESAT) were compared to simultaneous blood flow values measured by laser Doppler flowmetry and venous oxygen-saturation (SVO/sub 2/) determined from direct blood gas analysis. The regression for flow provided a zero p-value, a variance S=17.57 and F statistic=50.49. The residuals vs. fits plots suggest that the current model would underestimate values below the mean and overestimate those above the mean. An improved regression model for SvO/sub 2/ provided a zero p-value, a variance S=14.1 and F statistic=4.26. Two different neural networks were implemented for flow and oxygen saturation. Both networks "tracked" their values closely and with low cycle errors. Neural networks are powerful tools for evaluation of rapidly changing, variable environments.

We propose a new computerized system to accurately guide laser shots to the diseased areas within the retina based on predetermined treatment planning. The proposed system consists of a fundus camera interfaced to a computer that allows real-time capturing and processing of a sequence of images at video frame rates. The first image in the sequence is used as a reference for manual treatment planning. A new segmentation technique was developed to discern the blood vessel tree in which we extract the boundaries of the wide vessels as well as the whole lumen of smaller ones. The core of wide vessels is then obtained by correlating the image with a 1-D Gaussian filter in two perpendicular directions and thresholding the result. A fast registration technique is proposed to automatically track the motion of segmented areas within the subsequent images in the sequence. In this technique, points satisfying certain developed significance conditions are chosen as landmarks in the reference frame. Using the extracted set of corresponding points in the subsequent image frames, we can accurately align image frames to compensate for eye movements and saccades. The new technique has the potential to significantly improve the success rates in this type of treatment.

If detected early, breast cancer can be treated with better patient outcomes and significantly lower costs. From recent (1998) retrospective breast cancer studies, in approximately half of missed cases, a minimal sign was already visible on a prior mammogram. Using information technology such as spatial dendrograms (stealth-related technology) and repartment hierarchical identification (successive information peeling), difficult cases of spiculated and stellate tumours can be identified. The techniques are robust to noise and can reveal various layers of biophysical and biomedical differences in a tumour.