Development and Testing of a Prototype for 3D Radial Pulse Image Measurement and Compatible With 1D Pulse Wave Analysis

This study proposes the prototype of a wrist pulse acquisition device with three-dimensional pulse image (3DPI) obtained at the radial artery. The proposed prototype is designed with mechanical and electronic circuit hardware as well as the control and data acquisition software to have fourteen degrees of movement freedom against the sophisticated geometry of the wrist radial artery for accurate pulse-taking. Three robot fingers with nine degrees of movement freedom and a wrist fixer with two degrees of rotation freedom and one degree of manual movement freedom are designed to mimic the fine pulse-taking skill of a Chinese medicine physician. The customer-designed array sensors with $5\times5$ -1 sensing elements in an array are attached on the robot fingertips to get 3DPI with spatial information for direct translation of pulse feeling patterns in Chinese medicine texts, in addition to being compatible with 1D pulse wave analysis. To check reliability of the prototype, adopted is SphygmoCor with pulse data correlation. Consequently, the prototype passes safety tests and has high correlation of pulse data (waveform and peripheral augmented index) with those obtained by SphygmoCor. The automatic multi-depth pulse-taking with robot fingers attached array sensors on their tips can get acceptably repeatable pulse data with coefficient of variation less than 10%. By simultaneous pulse-taking at Cun-Guan-Chi positions, 3DPIs demonstrate great different stereoscopic waveshapes against almost unchanged single temporal pulse waveshape. Hence, the proposed prototype may be a useful tool to perform array pulse-taking for Chinese medicine pulse diagnosis and pulse wave analysis for cardiovascular examination.

SECTION I. Introduction

Pulse diagnosis has been generally applied in medical applications. For the acquisition of pulse waves with adequate quality and reliability, a variety of pulse diagnostic devices had been developed since the first, sphygmograph, invented in 1854 by Karl von Vierordt [1], with low resolution and precision due to its mechanization structure. Nowadays, the high-precision pulse diagnostic devices had launched, such as SphygmoCor, Arteriograph, Hem-9000AI, Vicorder, and Complior [2]–​[4]. Among them, SphygmoCor (Fig. 1) is considered the recommended device with thousands of peer-reviewed papers in clinical studies [5].

FIGURE 1.

A diagram of SphygmoCor pulse-taking.

Pulse diagnosis in Chinese medicine (CM) is also an essential diagnostic tool [6]. A well training Chinese physician presses the patient’s radial arteries in left or right wrist to read the pulse wave patterns by fingertip sensations for understanding the patient’s health condition. There are several finger-reading rules and lots of pulse wave patterns introduced in CM. Nan Jing (The Classic of Difficulties) strongly recommends three positions and nine indicators (TPNI) finger-reading rule for pulse-taking at the radial artery with about twenty pulse feeling patterns included [7]. In TPNI finger-reading rule, Chinese physicians press the radial arteries by three fingers on the Cun-Guan-Chi position at different displacements, Fu-Zhong-Chen (light-medium-heavy touch), for patient’s pulse wave patterns examination shown in Fig. 2. Due to the pulse wave pattern taken at different positions and displacements, a Chinese physician is able to obtain the patient’s health condition and provide medical treatment. Comparing Fig. 2 with Fig. 1, the pulse-taking by SphygmoCor and fingertips adopts the array pressure sensors; nonetheless, SphygmoCor takes the best single temporal pulse wave while CM physician takes three-dimensional pulse feeling patterns.

FIGURE 2.

A diagram of fingertip pulse-taking.

To measure the CM pulse feeling patterns, a series of wrist pulse acquisition devices or pulse diagnosis instruments (PDI) have been proposed and summarized in Table 1. Non-contact optical pulse-taking methods [8] even acquired spatial pulse information but faced the challenge of denoising and correlating optical measurement with pressure waves. Most of pulse acquisition devices similar to SphygmoCor only adopts single temporal pulse waves without spatial information to translate pulse feeling patterns [9]–​[15]. Some devices containing the linear array sensor got one-dimensional pulse information, pulse width, in addition to single temporal pulse waves [16], [17]. Several devices adopted array sensors to get three-dimensional pulse images (3DPI) with two-dimensional spatial information plus single temporal pulse waves, but with insufficient spatial resolution or low quality of 3DPI [18]–​[20].

TABLE 1 Review and Comparison of Wrist Pulse Acquisition Devices

In addition to our custom-designed array sensors for the increase of spatial resolution, the prototype designed by this study employs novel pulse-taking technology with high degree of mechanical movement freedom, which ensures the robot fingertips with array sensors placed on the middle artery fighting against the sophisticated geometry of a radial artery passing through Cun-Guan-Chi positions for the accurate pulse data acquisition. Three main contributions in comparison to the existing devices are summarized: (1) 14 degrees mechanical movement freedom fighting against the sophisticated geometry for the accurate pulse data acquisition, (2) the array sensor with $5\times5$ - 1 sensing elements is designed by this work to raise the spatial resolution in 7.2 mm $\times9.5$ mm, an area similar to the high-sensitive region of a human fingertip, (3) the quality of single temporal pulse waves got by the prototype is highly correlated with those of commercial SphygmoCor to ensure the accurate pulse data acquisition and thus be compatible with 1D pulse wave analysis (PWA).

The rest of the paper is organized as follows. In Materials and Methods section, the development principle, hardware, software, pulse-taking procedure, and signal processing and analysis are introduced. In Results section, safety tests, array data acquisition, repeatability test, correlation test, and 3D pulse imaging are presented. The final section discusses the experimental results and makes a conclusion.

SECTION II. Materials and Methods

A. Principle of the Proposed PDI Development

For radial pulse taking and signal analysis, inclusively to meet the rules of CM theory, the proposed prototype of the pulse diagnosis instrument (PDI) is designed according to the following four principles:

1. Using an array sensor on a robot probe or finger to acquire a 3DPI instead of a one-dimensional temporal pulse wave to obtain spatial information of pulse feeling.

2. The pulse-taking by the PDI can follow CM standard TPNI (three positions and nine indicators) by three robot fingers with automatic multi-depth control.

3. The PDI’s mechanical system has a multi-degree of movement freedom to mimic the sophisticated pulse-taking skill at Cun-Guan-Chi positions against the biodiversity of a human wrist in shape and size.

4. The PDI has general specifications for arterial pulse studies, including safety and repeatability, and the obtained pulse data highly correlated with those of the commercial instrument.

B. Hardware

The main hardware block diagram is shown in Fig. 3. Arduino Mega 2560 (Arduino, LLC, Italy), an open-source electronics prototyping board [21]–​[23], is embedded in the prototype to be a major control center, including a platform movement system, a sensor movement system, a motor homing system, and an assisted artery position system. The platform movement system and the sensor movement system are integrated into the actuator system, including 11 stepper motors in which 2 motors control platform movement in the Y and Z directions and 9 motors control three robot fingers movement in the X, Y, and Z directions (3 motors per finger). The motor homing system is designed to detect the position of sensor or platform if it reaches the origin point. A customized PPS sensor (Pressure Profile Systems, Inc., USA) is currently embedded for pulse data acquisition. A USB webcam and a low power laser are adopted for the operator to guide the robot fingers in the right position ready for automatical movement of three robot fingers on the radial artery. Additionally, a standalone wrist fixer system is designed to adjust the artery right facing against the robot fingers with 2 stepper motors, and to assist the laser positioning with 1 manual lead screw.

FIGURE 3.

The main hardware block diagram.

1) Actuator System

The primary function of the actuator system receives and decodes a command from the computer via USB communication to control the stepper motors for the movement of robot fingers with array sensors on their tips. The actuator system consists of the following four parts:

1. The platform movement system. A platform on the PDI carries three robot fingers and nine motors (three motors per finger for X, Y, and Z directions). The platform movement system is responsible for the wide-range movement. Two 610 mNm high torque stepper motors (Minebea Company, Ltd., Japan) drive the platform to move in the Y and Z directions. The overall arrangement of the platform and sensor movement with 11 motors is shown in Fig. 4.

2. The sensor movement system. As seen in Fig. 5, there are three robot fingers with tactile array sensors on their tips. These three fingers are independently driven along the X-Y-Z axis to make the sensors reaching three positions (Cun, Guan, and Chi) precisely by nine fine-resolution (0.002 mm per step) stepper motors (Oriental Motor Company, Ltd, Japan). Due to the big size of stepper motors, the left and right motors of up-down movement are arranged with 45 degree included angle with the middle motor.

3. The motor homing system. The homing system is designed to detect the robot fingers or platform whether it reaches the motor origin point. The origin point is the starting point as the instrument is started or reset to provide high-accuracy reference for positioning calibration. The homing system adopts 2 limit switches for platform position calibration and 9 photo interrupters for sensors position calibration (Fig. 6).

4. A 24VDC, 10A power supply is selected in the system. A power converter is designed due to the different power supply requirements of modules. The schematic and PCB of the power converter are shown in Figs. 7 and 8. In the schematic, U1 and U2 convert 24VDC to 12VDC and 5VDC. D1 and D2 are freewheeling diodes to eliminate voltage spikes. Capacitors C1 to C4, parallel with the input and output of U1 and U2, can lower down voltage ripples. In Fig. 8, three different socket types (upper panel) are adopted for a Poka-yoke design, and a DC brushless fan is used to cool down components’ temperature (lower panel).

FIGURE 4.

Top view of the platform movement system with two stepper motors S4 (Y and Z), and the sensor movement system with nine stepper motors S1-S3 (X, Y, and Z per robot finger). X and Y for horizontal movement while Z for up and down movement.

FIGURE 5.

The industry design of 3 robot fingers with nine stepper motors (upper panel) and the independent movement of robot fingers to radial artery (lower panel).

FIGURE 6.

An example of the motor homing system with photo interrupter modules installed for S1-Z, S2-Z, and S3-Z motors.

FIGURE 7.

The schematic of the power converter.

FIGURE 8.

PCB layout (upper panel) and installation (lower panel) of the power converter.

2) Tactile Array Sensor

A previous study [20] found that, compared to existing commercial products, the commercial $4\times6$ array sensor manufactured by pressure profile system company (PPS) was the best choice for pulse measurement with good performance in linearity, repeatability, and sensitivity. However, the PPS $4\times6$ array sensor in 8 mm $\times12$ mm area may not be specially designed for pulse diagnosis at the radial artery. The sensor output is connected at the wide side with 4 sensing elements to be only installed on the robot fingertips facing against the longitudinal direction of a radial artery. Consequently, only $4\times4$ sensing elements in $4\times6$ array sensor can effectively measure the radial pulse data. It is also expected to use odd number of sensing elements in an array, so the central element can be located on the center of a radial artery to get symmetric data acquisition.

Therefore, we mimic the human fingertip structure to order PPS array sensor with $5\times5$ – 1 sensing elements in 7.2 mm $\times9.5$ mm area in Fig. 9, in which one sensing element is absent on the corner whose pulse data can be obtained by interpolation from pulse data of the adjacent sensing elements. The rectangular geometry of a sensing element in 1.25 mm $\times1.71$ mm is arranged to have the same number of sensing elements in the longitudinal and transverse directions of a radial artery for the symmetric data acquisition. It should be noted that the sensor output is connected at the long side of the array sensor that can be easily attached to a robot fingertip to face against the longitudinal direction of a radial artery. The specifications of the customized PPS array sensor include a full-scale range of 300 mmHg, a sensitivity of 0.5 mmHg, a scan rate of 50 Hz, a temperature range of −20 to 100 °C.

FIGURE 9.

The customized PPS sensor with the sensing elements rearranged to $5\times5$ – 1 on the sensor tip.

Given the PPS sensors on the robot fingertips that touch the human radial artery, we use a standalone Bluetooth module (D710, PPS) driven by a 5-volt lithium battery to transmit pulse data wirelessly to the computer (Fig. 3). The electronic isolation by wireless transmission physically prevents the transfer of direct current (dc) and unwanted alternating current (ac) in the PDI to hurt people or contaminate pulse data.

3) Assisted Positioning System

The assisted positioning system including a 100mW laser and a 720HD USB webcam (Fig. 10) is used to help move robot fingertips touching Cun-Guan-Chi positions accurately. The laser is installed in front of the platform with a fixed distance from the origin of the middle robot finger for Guan position. The platform can be adjusted to let the laser beam pointing at the subject’s Guan, on which the Guan robot finger is automatically moved and placed. Besides, the webcam can be adjusted with a wide angle to view the positioning point easily.

FIGURE 10.

The Assisted Positioning System with a laser and a webcam.

4) Wrist Fixer

The wrist fixer with arm support and palm rest is a standalone system designed to adjust the artery right facing against the robot fingers. The controller of the wrist fixer is an Arduino Nano board based on the ATmega328 processor controlling 2 Minebea stepper motors with 4 manual buttons on the front panel of the prototype. As shown in Fig. 11, when the left or right arm is put at the wrist fixer fixed by Velcro straps, the X-axis rotation of the arm support is to make the measurement artery area right against the sensors. A Y-axis rotation of the palm support is to maintain the subject’s radial artery at the same height for accurate measurement [24]. Additionally, a manual lead screw is used to move the wrist fixer along the X-axis for the laser easily pointing at Guan position of the artery.

FIGURE 11.

Picture of the wrist fixer.

Fig. 12 shows the prototype in whole. An emergency button is set on the front panel to directly lift the platform by hardware control for stopping the pulse-taking in case of overpressure. Table 2 shows the prototype specifications, in which the current prototype is relatively large ($62\times 47\times41$ cm³) due to more than a dozen of driver modules for stepper motors. The platform is moved with large-scale steps while the sensor movement system achieves a fine- precision movement in $67.0\times 47.3\times30.0$ mm³ (or 95.1 cm³) range for accurate and flexible measurement.

TABLE 2 The Prototype Specification

FIGURE 12.

Picture of the prototype.

C. Software

We programmed the system software with LabVIEW, which is widely used in the development of the experimental biomedical equipment [25]. The advantages of LabVIEW include but not limited to flexibility, scalability, as well as a friendly graphical user interface (GUI) [26], [27]. The GUI contains three main units, as shown in Fig. 13:

FIGURE 13.

The main user interface. (a) Find patient interface, (b) new patient interface, (c) measurement interface with hardware control and data acquisition, (d) playback interface.

1) Database Unit

Via database unit, the personal information can be created and edited in Fig. 13(a) and 13(b).

2) Measurement Unit

This is the main unit to execute the hardware control and data acquisition shown in Fig. 13(c). The prototype platform driven by 2 stepper motors and the robot fingers driven by 9 stepper motors can be moved to the assigned positions by dragging 2 knobs and 9 slides on the right panel of GUI or clicking their corresponding buttons on the PC keyboard, with the assistance of the real-time cam (left top panel) and 3DPI window (left bottom panel). For automatic multi-depth measurement needs, the initial position, end position, and the number of pulse-taking between these two positions can be set in the depth sets panel (middle top panel).

3) Playback Unit

This unit supports data playback simultaneously in both 3DPI and 1D pulse waves with play, pause, and stop buttons in the left top corner in Fig. 13(d).

In addition, we introduced pressure limit control during the pulse-taking procedure for safety. According to the national standard Pulse Graph Acquisition Device for Traditional Chinese Medicine (YY/T 1489-2016, China), the human wrist is considered to be safe by pushing pressure under 660 mmHg (or 88 kpa). Empirically, as the pressure approaches 360 mmHg, the robot fingers almost reach the bone under the radial artery. So, the maximum pressure currently applied by the control system is limited at 360 mmHg for the stepper motor movement, much lower than the 660 mmHg recommended by the national standard. As the maximum is reached, the robot fingers stop pressing down.

D. Pulse-Taking Protocol

A subject is arranged to sit in a chair in a quiet room for rest at least 10 minutes to make sure that he/she calms down [28]. As shown in Fig. 14, the subject and the operator sit in suitable relative positions, where the subject can be comfortably tested on his/her wrist with height of his/her heart, and the operator can conveniently operate the computer and manage the measurement process. Then, the SphygmoCor (model EM3, AtCor Medical, Inc., Australia) is employed to measure the single pulse waves at Guan position. Next, the PDI is used to take the array pulse waves following the recommended protocol shown as the flowchart below:

1. Put the left or right hand of a subject into the wrist fixer, then fix the hand and the wrist with the velcro straps.

2. Keep the wrist parallel to the sensor plane by adjusting the wrist fixer.

3. Search and mark three positions - Cun, Guan, and Chi - by feeling the pulses manually.

4. Focus the laser point on the mark of Guan and press the laser tracking button to move the platform with Guan robot finger right on the top of the mark.

5. Press down on Guan and slightly adjust the position of Guan robot finger for apparent pulse beating shown on the screen display.

6. * After the Guan position is ready for pulse data record, search and press Cun and Chi by moving their corresponding robot fingers manually on the marks.

7. * Raise up three robot fingers and stop them at 3 mmHg static pressure. Record the positions of three robot fingers as their initial depths.

8. * Press down three robot fingers until the pressure reaches a maximum of 300 mmHg. Record the positions of three robot fingers as their final depths.

9. * Set the step number 15, as an example, so the initial and final depths are the 1st and 15th step separately. Push data record button to automatically record Cun-Guan-Chi pulse data at 15 depths with 10 seconds per step.

FIGURE 14.

Pulse-taking using our prototype.

*: The step number and recording period per step can be set freely according to different experimental designs.

E. Signal Processing and Analysis

The SphygmoCor yields a single pulse wave by choosing the best one in an array, however, the proposed PDI yields 25 pulse waves per position (Cun, Guan, and Chi on left and right hands) by taking all pulse data in an array. To compare the pulse data quality between the SphygmoCor and PDI, we choose the single pulse with the highest amplitude of array pulse data in every position for the prototype by the developed algorithm.

The data cycle length of pulse waves by the PDI and SphygmoCor are not equal due to the different sampling rates between PDI (50 Hz) and SphygmoCor (128 Hz) and the heart rate variability. In Matlab, both waveforms were firstly conducted with cubic spline interpolation to get the same cycle length and then calculate the Pearson correlation coefficient.

The augmented index (AI) is an important metric for the evaluation of vascular sclerosis and can be divided into the peripheral (pAI) and central (cAI) ones [29]. Since both PDI and SphygmoCor acquire pulse waveforms at the radial artery near wrist, we merely focus on the comparison of pAI. The definition of pAI for a radial pulse waveform is the ratio of the percussion peak (P₁) divided by the tidal peak (P₂) resulting from a reflective wave.

P₁ can be easily found as the local maximum and minimum values of a pulse cycle. P₂ must follow the inflection point of the descending phase, which is detected by the local maximum of the second derivative of the pulse wave [30]. So, we can find P₂ after the inflection point at the local maximum if it is an explicit peak, while it is found at a zero-crossing point in a wavelet wave analysis if it is an implicit peak [31].

SECTION III. Results

A. Safety Tests

The prototype to be a medical instrument must consider the safety issue seriously. To make sure the experimental safety, performed was a series of safety tests (Table 3), including electrical leakage tests and electromagnetic compatibility tests in Compliance Certification Services Inc. (CCS), an officially certified testing agency in Taiwan. All safety tests were passed to ensure the safety of subject test.

TABLE 3 Safety Tests for the Proposed PDI

B. Array Data Acquisition

This study received the human trial approval from the institutional review board at the National Cheng Kung university hospital (Approval Number: B-ER-103-263). Personal data collected from 58 subjects with the SphygmoCor and the proposed PDI successively were examined and analyzed (Table 4).

TABLE 4 Subject Characteristics

Three batches of simultaneous palpation pulse waves on Cun, Guan, and Chi are generated after the robot fingers pressed from lightly touching skin to the deep in 15 steps. The raw data with time and pressure information is plotted in two-dimension form by MATLAB, which includes 25 channels per finger. A Guan data is demoed in Fig. 15 with the full view in the upper panel and a zoom-in view in the lower panel. Each channel data has a stair-stepping DC level due to the different contact pressure with a little perturbation by respiration. Note that the pulse amplitude enlarges in the first half steps then decreases afterward due to the radial artery is over-compressed, which is consistent with the previous pulse depth study [14]. The data, as yet, can be considered as a group of 1D pulse waves in similar shapes but with different amplitudes.

FIGURE 15.

Simultaneous palpation pulse waves in Guan position with the overall view (upper panel) and zoom-in view (lower panel).

C. Repeatability Test

Table 5 shows that the pulse amplitude of two subjects measured at three positions of their left radial arteries near wrists (L1, L2, and L3 mean left Chi, Guan, and Cun). There were five tests per subject in similar environments in three days by the same operator. The pulse amplitude is acquired at 100 mmHg robot finger pressure. The data is shown in mean and SD (Table 5). In five tests, there is one test a little bit far away from the average value, resulting from the robot fingers not covering the whole artery very well. So, coefficient of variation (CV) values span from 4.5% to 25.1% for five tests per position; nevertheless, most CV is less than 10% for four tests per position if a far-deviation data is deleted. It is recommended to conduct successive two measurements and make sure these two-measurement data close enough to ensure good repeatability. If not, conduct the third measurement carefully by placing the radial artery on the center of the robot fingers and choose one of two close data as the measurement data. In addition, it is an interesting feature that the pulse amplitude at L2 was higher than that at the other two positions in these two cases, meeting with the previous study [32].

TABLE 5 Pulse Amplitude at Three Positions

D. Correlation Test

A flow chart of correlation test shown in Fig. 16 includes waveform correlation test and pAI correlation test with all 58 subjects between PDI and SphygmoCor. Data were processed with Matlab, and the statistical analysis was performed with SPSS.

FIGURE 16.

The flow chart of correlation testing.

The waveform cross correlation between pulse waves of left (L1-L3) and right (R1-R3) radial arteries by the PDI and L2 by SphygmoCor is listed in Table 6 and Fig. 17. Obviously, the pulse shape of the PDI has high correlation (mean greater than 91%) with that of SphygmoCor.

TABLE 6 Correlation of Waveform

FIGURE 17.

Box-plot of the waveform correlation.

In addition to waveform correlation, pAI is calculated for both pulse waves at L2 measured by the prototype and SphygmoCor. In Table 7, it is shown that both pAI groups have a mean difference less than 2% within a measurement tolerance of biodata. Subsequently, there is a close correlation shown in Fig. 18 ($\text{R}^{2} = 0.71$ ; regression equation: Y = 0.84x+10.36) [3].

TABLE 7 Correlation of pAI

FIGURE 18.

Analyses of regression correlation and Bland-Altman plot for pAI estimated by SphygmoCor and the proposed PDI. Correlation with regression equation: Y = 0.84x+10.36 (upper panel). Bland-Altman plot showing agreement between two groups of pAI values (lower panel).

E. 3D Pulse Imaging

Shown in Fig. 19 is a group of 3D pulse images at three positions (Cun, Guan, and Chi), filtered using a 0.1-20 Hz bandpass Butterworth filter and formed by a cubic interpolation algorithm in Matlab. The 1D temporal pulse waves at three positions are listed on the top, and shown below are five stereoscopic (left side) and planar (right side) 3DPIs with normalized sensing length, width, and pulse amplitude for concise illustration at the timing marked on the 1D temporal pulse waves. The 1D pulse waves at Cun, Guan, and Chi are arranged from right to left to compromise with the blood flow direction from Chi to Cun. It is worth noting that the 1D pulse waves at three positions are similar in shape, while their corresponding 3DPIs have significant different image shape. So, 3DPI directly translates pulse feeling patterns into quantitative image data for CM pulse diagnosis [33].

FIGURE 19.

3D pulse images with normalized sensing length, width, and pulse amplitude demoed on five marked timings from their 1D pulse waves in a heartbeat (No.1 - cycle start; No.2 - middle timing of No.1 and 3; No.3 - percussion peak; No.4 - tidal peak; No.5 - dicrotic peak). Stereoscopic (left side) and planar (right side) 3DPIs shown in (a) Chi, (b) Guan, and (c) Cun.

SECTION IV. Discussion and Conclusion

For the experimental safety, in addition to a series of safety tests in CCS, an emergency button is added to raise up the PDI platform for releasing the pulse-taking pressure. Also, the pulse-taking pressure limit control is introduced in the software.

The prototype can get acceptable repeatability with CV less than 10% within allowable tolerance possibly caused by diurnal physiological variation [34], [35], as the radial artery is pushed as much close at the center of the robot fingers to ensure the correct measurement of pulse amplitude. It is strongly recommended to measure the pulse data twice and check if they are close enough to ensure successful measurement for good repeatability.

In comparison to the commercial tonometry, the pulse waveform and pAI of the prototype have high correlation with those of SphygmoCor. SphymoCor adopts an array sensor with four elements while only produces one single temporal pulse wave as the measurement data [3], [36]. This is different from the prototype using every sensing pulse wave in an array to form a 3DPI.

A radial artery passes through the Cun-Guan-Chi area with sophistical geometry, not a simple straight line. Also, it is challenging to keep the fingers touching right on the radial artery, even for an experienced CM physician. Therefore, a wrist pulse acquisition device must have a sufficient number of movement dimensions to place the robot fingers right on the radial artery. Especially for 3DPI, every sensing element in the array must touch the artery firmly to get the accurate pulse data. It is required to have at least 11 movement dimensions in which 9 movement dimensions are designed for three robot fingers (3 dimensions per finger) and 2 movement dimensions are for wrist and palm rotation to adjust the artery right facing against the robot fingers. With such a high degree of movement freedom (at least 11 mentioned above), it makes possible to overcome the sophistical geometry of the radial artery in the Cun-Guan-Chi area for accurate 3DPI acquisition. That is why the prototype has 14 degrees (13 motor-driven and 1 manual) of pulse-taking freedom.

For pulse diagnosis, the relationship between arterial stiffness and the spatial geometric characteristics of 3DPI has been found [37], which enhances PWA extending from 1D to 3D space. Besides, an ingestion test revealed that 3DPI at different positions and depths of the radial artery might independently respond to a variety of physical conditions [38], consistent with CM texts [33]. In a word, against the almost unchanged waveshape of a single temporal pulse wave, 3DPI does change its spatial waveshape in response to the different sensing positions, which reflects the pulse feeling patterns in CM. In Fig. 19, it is a normal 3DPI at Chi, while it looks like a slippery 3DPI with 2 or 3 peaks on top of 3DPI at Guan [33]. At Cun, it is very difficult to see a clear spatial waveshape. Therefore, it is strongly recommended to conduct a quantitative analysis of 3DPI instead of a single pulse wave for CM pulse diagnosis.

In sum, the proposed prototype has sufficient pulse-taking degrees of freedom with array sensing 3DPI to duplicate the pulse feeling patterns and mimicking CM physician’s pulse-taking skill for accurate pulse data acquisition, which is also compatible with 1D pulse wave analysis. Although the customized array sensor introduced in this study is the best commercially available for measuring 3DPI, its performance such as the sampling rate needs to be further improved in the near future. Nonetheless, the prototype must take pulse data twice to ensure high repeatability, so it needs further improvement to avoid a long recording time.

ACKNOWLEDGMENT

Many thanks must go to Research Assistant H.-E. Fang for his assistance in the industrial design of the prototype. (Bo Peng and Ching-Hsing Luo are co-first authors.)