Keywords

1 Introduction

Use of ionizing radiation (x-rays) to produce radiographs is the most common procedure for doctors to differentiate a wrist injury into a sprain or a fracture. Each wrist radiograph costs around £60 in the United Kingdom and almost 60 percent of those radiographs show no fracture. Thus, a non-invasive technique is very desirable to minimize both radiation exposure and cost.

We have used a small, computer run, hand-held system that generates vibration in the bone under examination, and then captures generated vibration signals by a piezoelectric sensor, which then are analyzed by applying signal processing methods. Various technological systems in both industry and medicine apply either Fast Fourier Transform (FFT) or Wavelet Coefficient extraction to process acquired signals. In this study, we have applied both methods on signals obtained from human wrist bones and then compared the results to find out which method is more accurate to screen for fracture.

Vibration analysis is an established method used in industry for monitoring or diagnosing faults in electric machines by means of signal analysis and signal processing. Mechanical vibrations cause one or more inputs and outputs of a system to change over time and the signals produced are then evaluated for type and severity of defect [1]. Studying the vibration effects, fault detection and capability of this method of machine maintenance, is an on-going process [2, 3].

By comparison, vibration analysis in the medical field is still in its infancy. In 1932, Lippmann percussed his finger to cause vibration, used his stethoscope to sense it and explained the benefits for using the method in clinical diagnostics. He suggested that analysis of percussion appeared more accurate than x-ray for the diagnosis of fractures [4]. Some researchers developed a finite model of a dry tibia for predicting its natural frequencies of vibration [5]. Nokes published a major study where he proposed the idea to monitor various pathological and trauma-induced conditions by applying a stimulus to bone [6]. Experiments demonstrated that vibration induced by mechanical means can indicate mechanical properties of bones, such as density, biological state, etc. [7]. The influence and results of low-frequency vibrations in the lower limbs have been accurately determined for the analyses based studies [8]. The effects of skin and muscle on the bone’s low frequency vibration signals have been demonstrated [9]. We have previously used samples of frequency spectra from adults with and without wrist injuries, to cluster analyze into fractured and sprained bones [10].

Application of vibration analysis is still developing. Additionally, studies are being conducted to find out if its advantages outweigh the disadvantages. However, in the context of health care, the practical applications of vibration analysis are in the near rather that distant future.

2 Smart Systems

Smart systems are defined as devices that incorporate the functions of sensing, actuation, control and adaptation. They can describe and analyze situations, and take decisions based on the available data in a predictive or adaptive manner, thereby performing smart (intelligent) actions [11].

Smart systems are found in many applications such as automotive, manufacturing, environment, energy and health care. These systems have the capability to solve very complex problems, including taking over human cognitive functions [12].

In the case of health care and medical applications, smart systems are mainly used to improve the quality of diagnosis and treatment. The system described in this paper shows the utilization of a miniature smart hardware device to produce and procure energy through a medium. The first prototype of the device has been tested to find correlation between vibration frequency and bone density, and has been developed by the Medical Electronics Engineering Group at Sheffield Hallam University’s Centre for Automation and Robotics Research facility [13]. The device has been then modified and our group now uses it to differentiate between wrist fracture and strain. The data acquired from this method are stored and analyzed using complex programmed routines through specific software.

2.1 System

Hardware:

The device comprises of the following parts.

  • Electrical Vibration Inducer – A sample rate of 50000 samples per second makes the inducer move that produces the impact. Squared signals with a duty cycle of 50 percent at maximum, a frequency of 1 Hz, and an offset of 2.5 were selected for generation of the signals during the impacts.

  • Piezoelectric Sensor – The sensor, used for vibration/impact sensing has broad bandwidth, high sensitivity, light weight and low cost features.

  • NI myDAQ Assistant – This was used for frequency generation which causes the inducer to initiate and consecutively record the vibration response via the sensor.

  • 12 V Battery – This battery was added to remove the portability and electric shocks issues.

  • Control Circuit – A small Printed Circuit Board (PCB) was built which consists of amplifier, filter, and power divider to control the circuit for minimal battery power consumption.

Software:

LabVIEW and MATLAB.

The fundamental principle behind using LabVIEW is for its Integrated Development Environment (IDE) and the Graphical User Interface (GUI), which were used to develop the data acquisition feature for this device. MATLAB was used to process and analyze the acquired data. LabVIEW was designed in such a way that a NI myDAQ Assistant can generate signals to run the inducer and also acquire the corresponding data; in other words, both the generation and acquisition modes of the DAQ Assistant were utilized. Implementation loops were added to create mechanized applications to reduce manual tasks and enhance repeatability. MATLAB is an efficient tool for signal processing and that had been used to perform both FFT and WD.

Thus the system developed for this research, is undeniably a Smart System (Fig. 1).

Fig. 1.
figure 1

Process flow chart

Full size image

3 Research Methodology and Results

Adult patients, aged between 20 to 70 years old, attending the Emergency Department of a local hospital, following a wrist injury, volunteered to take part in this study. Vibration data were collected from both injured and uninjured wrists of six patients and processed blinded to radiographic findings. This means that at, the time the data were produced and analyzed, it was not known if the injured wrist had a sprain or a fracture. They all sat on a chair during recording. Vibration caused by mechanical excitation, was induced by impacting the most prominent bone in the elbow. The vibration responses were recorded by placing a piezoelectric sensor which has a maximum operating frequency, on a bony prominence at their wrists. The device was intended to induce light vibrations onto the bones for sensing the signals. The vibration inducer had been designed in such a way that its strength can be controlled, if required. All of the volunteers stated after participating in the study that they did not feel any pain due to the vibration induction.

10 vibrations were induced over 10 s in total and each response was recorded at a rate of 50000 samples per second. These data were stored in the computer and then processed by Wavelet Decomposition (WD) and Fast Fourier Transform (FFT).

  1. I.

    The coefficients were extracted, after decomposition of the acquired data by Level 5 filtration. These were derived by a scaling function “Daubechies 5 (db5)” wavelet family. Two degree polynomials were plotted to find the best fitted curve among the number of coefficients derived from the WD. The equation for each of the polynomials was also calculated.

  2. II.

    The bone vibration recorded from six patients was converted to a magnitude frequency spectrum by application of FFT processing.

Figures 2 and 3 display the results produced by FFT and WD.

Fig. 2.
figure 2

Filtered vibration response and magnitude spectrum by FFT.

Full size image
Fig. 3.
figure 3

Best fitting two-degree polynomial among 18 wavelet decomposed coefficients.

Full size image

The frequencies produced by the FFT and the constants derived from the equation produced by the best fitted polynomial curve, were classified using the “dendrogram” (a hierarchical cluster tree diagram) to show which groups are similar. The “dendrogram” was plotted using a linkage of Euclidean distance between each data set. The diagram produced by both FFT and Wavelet Decomposition was similar and is illustrated in Fig. 4.

Fig. 4.
figure 4

Both FFT and wavelet decomposed coefficients produced the same results by the dendrogram.

Full size image

These data were then analyzed using Fuzzy C-means to find the membership percentages required for each data set to belong to a particular cluster. Although similar classes can be observed from the dendrogram, Fuzzy C-means was performed to ascertain the suitable percentage required for each group. The results of cluster analysis are shown in Tables 1 and 2.

Table 1. Fuzzy C-means cluster analysed results after fast fourier transformations
Full size table
Table 2. Fuzzy C-means cluster analysed constants extracted from equation of the best-fitted polynomial among the wavelet decomposed coefficients.
Full size table

4 Discussions

Tables 1 and 2 clearly reflect the techniques used in this study.

WD produced comparatively similar results to FFT, while using both the dendrogram and Fuzzy C-means. However, there are some disadvantages and unreliable results of using the WD technique:

  • WD provides both Frequency and Time related information which makes it difficult to use; as the design which induced 10 vibrations must produce vibration responses very similar to each other over the 10 s in order to ensure that results can be processed further.

  • In order to make sure the above reason does not affect the accuracy of the data collected, the first and last few sets of data are left out when doing the analysis.

  • In addition, there are a sizable number of steps before reaching the concluding stage of Fuzzy C-means cluster.

On the other hand, the frequencies produced by the FFT can easily be fed into the Fuzzy C-means code already built, to produce membership values. Fewer steps were involved in building the FFT codes.

5 Conclusion and Further Works

Both FFT and WD techniques produced almost all correct results. However, the FFT is a hassle free and accurate process with less number of steps for future use. The efficiencies are to be checked by a larger volume of data. The discomfort level experienced by each patient for both wrists was recorded and every participant felt that the vibrations would not be uncomfortable to children aged between 0 and 15 years old. It takes less than 5 min to carry out the procedure so it is not a time consuming method. The study is being continued at a local Children’s Hospital and aims to recruit a larger number of participants to test the accuracy of the techniques. More signal processing techniques are likely to be assessed once a larger volume of data is available.