ABSTRACT
The process of diagnosing tremor patients often leads to misdiagnoses. Therefore, existing technical methods for analysing tremor are needed to more effectively distinguish between different diseases. For this purpose, a system has been developed that classifies measured tremor signals in real time. To achieve this, the hand tremor of 561 subjects has been measured in different hand positions. Acceleration and surface electromyography are recorded during the examination. For this study, data from subjects with Parkinson's Disease, Essential Tremor, and physiological tremor are considered. In a first signal analysis feature extraction is performed, and the resulting features are examined for their discriminative value. In a second step, three classification models based on different pattern recognition techniques are developed to classify the subjects with respect to their tremor type. With a trained decision tree, the three tremor types can be classified with a relative diagnostic accuracy of 83.14%. A neural network achieves 84.24% and the combination of both classifiers yields a relative diagnostic accuracy of 85.76%. The approach is promising and involving more features of the recorded time series will improve the discriminative value.
Subject(s)
Essential Tremor , Parkinson Disease , Electromyography/methods , Essential Tremor/diagnosis , Hand , Humans , Parkinson Disease/diagnosis , Tremor/diagnosisABSTRACT
Dedicated research is currently being conducted on novel thin film magnetoelectric (ME) sensor concepts for medical applications. These concepts enable a contactless magnetic signal acquisition in the presence of large interference fields such as the magnetic field of the Earth and are operational at room temperature. As more and more different ME sensor concepts are accessible to medical applications, the need for comparative quality metrics significantly arises. For a medical application, both the specification of the sensor itself and the specification of the readout scheme must be considered. Therefore, from a medical user's perspective, a system consideration is better suited to specific quantitative measures that consider the sensor readout scheme as well. The corresponding sensor system evaluation should be performed in reproducible measurement conditions (e.g., magnetically, electrically and acoustically shielded environment). Within this contribution, an ME sensor system evaluation scheme will be described and discussed. The quantitative measures will be determined exemplarily for two ME sensors: a resonant ME sensor and an electrically modulated ME sensor. In addition, an application-related signal evaluation scheme will be introduced and exemplified for cardiovascular application. The utilized prototype signal is based on a magnetocardiogram (MCG), which was recorded with a superconducting quantum-interference device. As a potential figure of merit for a quantitative signal assessment, an application specific capacity (ASC) is introduced. In conclusion, this contribution highlights metrics for the quantitative characterization of ME sensor systems and their resulting output signals in biomagnetism. Finally, different ASC values and signal-to-noise ratios (SNRs) could be clearly presented for the resonant ME sensor (SNR: -90 dB, ASC: 9.8×10-7 dB Hz) and also the electrically modulated ME sensor (SNR: -11 dB, ASC: 23 dB Hz), showing that the electrically modulated ME sensor is better suited for a possible MCG application under ideal conditions. The presented approach is transferable to other magnetic sensors and applications.
Subject(s)
Heart , Magnetic Fields , Magnetics , Signal-To-Noise RatioABSTRACT
Magnetoelectric (ME) sensors with a form factor of a few millimeters offer a comparatively low magnetic noise density of a few pT/Hz in a narrow frequency band near the first bending mode. While a high resonance frequency (kHz range) and limited bandwidth present a challenge to biomagnetic measurements, they can potentially be exploited in indirect sensing of non-magnetic quantities, where artificial magnetic sources are applicable. In this paper, we present the novel concept of an active magnetic motion sensing system optimized for ME sensors. Based on the signal chain, we investigated and quantified key drivers of the signal-to-noise ratio (SNR), which is closely related to sensor noise and bandwidth. These considerations were demonstrated by corresponding measurements in a simplified one-dimensional motion setup. Accordingly, we introduced a customized filter structure that enables a flexible bandwidth selection as well as a frequency-based separation of multiple artificial sources. Both design goals target the prospective application of ME sensors in medical movement analysis, where a multitude of distributed sensors and sources might be applied.
Subject(s)
Benchmarking , Movement , Motion , Noise , Signal-To-Noise RatioABSTRACT
The knowledge of the exact position and orientation of a sensor with respect to a source (distribution) is essential for the correct solution of inverse problems. Especially when measuring with magnetic field sensors, the positions and orientations of the sensors are not always fixed during measurements. In this study, we present a processing chain for the localization of magnetic field sensors in real time. This includes preprocessing steps, such as equalizing and matched filtering, an iterative localization approach, and postprocessing steps for smoothing the localization outcomes over time. We show the efficiency of this localization pipeline using an exchange bias magnetoelectric sensor. For the proof of principle, the potential of the proposed algorithm performing the localization in the two-dimensional space is investigated. Nevertheless, the algorithm can be easily extended to the three-dimensional space. Using the proposed pipeline, we achieve average localization errors between 1.12 cm and 6.90 cm in a localization area of size 50cm×50cm.