- Email: [email protected]
A precise and accurate acupoint location obtained on the face using consistency matrix pointwise fusion method
Online Submissions: http://www.journaltcm.com [email protected]
J Tradit Chin Med 2015 February 15; 35(1): 110-116 ISSN 0255-2922 © 2015 JTCM. All rights reserved.
METHODOLOGY TOPIC
A precise and accurate acupoint location obtained on the face using consistency matrix pointwise fusion method
Yang Xuming, Ye Yijun, Xia Yong, Wei Xuanzhong, Wang Zheyu, Ni Hongmei, Zhu Ying, Xu Lingyu aa Yang Xuming, Ye Yijun, Xia Yong, Acupuncture and Tuina College, Shanghai University of Traditional Chinese Medicine, Shanghai 201203, China Wei Xuanzhong, Wang Zheyu, School of Electronic Information and Electrical Engineering, Shanghai Jiaotong University, Shanghai 200240, China Ni Hongmei, Basic Medical College, Shanghai University of Traditional Chinese Medicine, Shanghai 201203, China Zhu Ying, Ophthalmology Department, Shanghai of Traditional Chinese Medicine Hospital, Shanghai 200071, China Xu Lingyu, College of Computer Engineering and Science, Shanghai University, Shanghai 200072, China Supported by the Key Program of State Administration of Traditional Chinese Medicine of China: the Science of Acupuncture and Moxibustion (No. ZYSNXD-CC-ZDXK-07) Correspondence to: Prof. Yang Xuming, Acupuncture and Tuina College, Shanghai University of Traditional Chinese Medicine, Shanghai 201203, China. [email protected] Telephone: +86-21-51322434; +86-18916005529 Accepted: February 27, 2014
the general weight calculation. First, we corrected each expert of acupoint location system error itself, to obtain a rational quantification for each expert of acupuncture and moxibustion acupoint location consistent support degree, to obtain pointwise variable precision fusion results, to put every expert's acupuncture acupoint location fusion error enhanced to pointwise variable precision. Then, we more effectively used the measured characteristics of different acupuncture expert's acupoint location, to improve the measurement information utilization efficiency and acupuncture acupoint location precision and accuracy.
Abstract
© 2015 JTCM. All rights reserved.
OBJECTIVE: To develop a more precise and accurate method, and identified a procedure to measure whether an acupoint had been correctly located.
Key words: Information fusion; Face recognition; Facial acupoint; Acupoint location
CONCLUSION: Based on using the consistency matrix pointwise fusion method on the acupuncture experts' acupoint location values, each expert's acupoint location information could be calculated, and the most precise and accurate values of each expert's acupoint location could be obtained.
INTRODUCTION
METHODS: On the face, we used an acupoint location from different acupuncture experts and obtained the most precise and accurate values of acupoint location based on the consistency information fusion algorithm, through a virtual simulation of the facial orientation coordinate system.
The effectiveness of acupuncture and its safe and scientific characteristics have been fully affirmed by the World Health Organization (WHO) and the National Institutes of Health (NIH), and acupuncture has become an important parts of the world's medicine.1,2 Many facial diseases,3 such as juvenile myopia, xerophthalmia, rhinitis and facial paralysis have shown significant treatment effects following acupuncture.4 Correct acupoint location is the first step of acupuncture therapy
RESULTS: Because of inconsistencies in each acupuncture expert's original data, the system error could not be modified using the characteristics of JTCM | www. journaltcm. com
110
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
and acupuncture treatment for facial diseases, because the facial capillaries and nerves are extremely rich, which is the key to the safety and curative effects of acupuncture and moxibustion. Three kinds of traditional acupoint location methods consist of basing location on anatomical landmarks of the surface of the body,5 measuring the distance between each part in Chinese anatomical inches (cun), and using the patient's fingers in proportion to a cun ruler. The cun measurement system is an essential component of traditional acupoint location methods used in acupuncture, dating back to 2900 BC, when pieces of jade resembling rulers were used for measurement.6,7 In 2002, an Australian scholar performed experiments8 on 72 participants selected from undergraduates of the Bachelor of Health Sciences in Acupuncture course, who repeatedly attempted to locate a fictitious acupoint (FP), with the resulting coordinates recorded. Significant differences were found between the two traditional methods (directional and proportional) and two contemporary methods (elastic and ruler). The 95% confidence ellipse for the ruler method had an area of 2.9 cm2, and that for the directional method had an area of 12.7 cm2. However, in terms of ease of use, most of the ratings for the traditional methods were 'comfortable', while almost half of the ratings for the two variant methods were 'uncomfortable'. Unfortunately, the two more precise contemporary methods were generally not well received; the two traditional methods were not with high precision, but which could comfortably provide with percived appearance to patients.9 In 2012, following the facial structure theory,10 Chinese scholars partitioned the facial features, using the Minimum Eigenvalue operator to detect the corner points of the features and the Log operator to detect the edges of the features; the comprehensive application of corner and edge information was used to localize each facial feature position. Finally, using the facial feature location as a reference coordinate, the facial acupoint was located, and according to the difference between the left and right point temperature of the face, the Baer facial acupuncture acupoint therapy was automatically selected. However, an abnormal target facial image could lead to a low accuracy rate of acupoint location. Because of the special nature of the face, in the metheds of "bone standard","finger cun measurement"and "anatomical landmark", the vast majority of acupoint used the "anatomical landmark" acupoint location method. An example is Chenqi (ST 1) location: on the face, directly below the pupil, in the depression of the infraorbital foramen.11 Here, we only discuss cases of acupoint location that used the "anatomical landmark" method. In short, the accuracy and precision of acupoint location methods have important implications for research. Treatment outcomes in clinical trials could be significantly altered if an inaccurate method of acupoint locaJTCM | www. journaltcm. com
tion was used. Practitioners must be able to locate on an acupoint with sufficient precision and accuracy to avoid stimulating nearly acupoints, since they might have effects other than those desired. At the same time, practitioners rely on different acupoints that have different physiological effects. Patient welfare, clinical efficacy, and valid research results depend on accurate and precise acupoint location. Moreover, different acupuncture prescriptions have the specific effect of treating different diseases.12 Therefore, we implemented an accurate and precise acupoint location learning method on the face, which was characterized by: establishing a facial orientation coordinate system with a virtual simulation, collecting acupoint location information from acupuncture experts, and using the consistency fusion algorithm for acupoint location, to get the most accurate and precise acupoint location values. The specific steps were as follows: (a) On the face, locating the acupoint and labelling it, then collecting front face photos and importing them into the computer; (b) In the computer's virtual environment, using local facial feature analysis and graphics and a neural recognition algorithm to determine the human facial organs and features, to build the coordinate origin and establish the coordinate system; (c) With multiple sources of information in time and spatial coordinates and the advantage of complementary and redundant information in common or joint operation, using the consistency information fusion algorithm, by fusing acupoint values from different acupuncture experts, to obtain precise and accurate coordinate values; (d) Establishing an acupoint database and constructing an interactive visualization learning interface, which lets the acupuncture practitioners who used the expert location values in the computer manipulation of the virtual learning environment establish acupoint location and prepare acupuncture prescription.
DATA ACQUISITION METHOD OF FACIAL ACUPOINTS Following the principles of Traditional Chinese Medicine (TCM), an acupoint location method was determined that suited the "anatomical landmark" method, collecting different acupuncture experts' acupoint location values to produce the best accuracy and precision values of each acupoint obtained via the consistency information fusion algorithm, providing a platform of facial acupoint location learning. Facial recognition technology to create a virtual simulation location system The specific methods were as follows. In a computer virtual environment, partial facial feature analysis and a graph and nerve recognition algorithm were used to determine the characteristics of the body's facial organs and features. Next, the origin of the coordinates was determined and the coordinate system was established. 111
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
Locating and identifying the acupoint on face Acquiring the front face photos Using the computer face recognition technology, to determine the human face organs and characteristic of the region, establishing coordinate system Getting acupuncture expert's acupoint location values By the technology of information fusion, calculating each acupuncture point "the best precise value" Establishing expert knowledge base of acupuncture acupoint Constructing digital learning interface of human-computer interaction Figure 1 One kind of precise facial acupoint location learning method flowchart
(a) Facial recognition procedure: to establish the facial image archives, facial images of each person were taken using a camera or were inputted from other photos to form the face image files, and were then transformed into Faceprint code that was stored in the archive base. A current Faceprint was generated from human face images taken by the camera, or from inputted pictures. The current Faceprint was compared with the Faceprint that was stored in the archive base. The Faceprint approach was based on the essential face characteristics and the facial shape. It could ignore light, skin tone, facial hair, hairstyle, glasses, facial expressions and posture change with strong reliability, so that it could accurately recognize the person from the pictures. The face recognition process, using common image processing software, could be completed automatically and continuously, in real time. (b) Establishment of the coordinate system: the algorithm was used to identify the face, and then the origin of a coordinate axis for the midpoint of the two eyes was identified, using the straight line between pupils for the X-axis. Because it was impossible to ensure that participants sat exactly in the same place, the X-axis of each photo coordinate system was not at the same level. Thus, the facial data were normalized to guarantee that they were comparable. In each photo, the person's own facial properties were used as a reference to establish the coordinate system. In practice, intended to the interpupillary distance (PD) 1/80 of a unit, to solve the problem that everyone's face is a different size.
rectangular features and the integral image. The algorithm was: (a) Using Haar-like features for testing: haar-like features: a region was characterized by means of facial detection under a sub-window. The pixels of the region were based on certain formula-calculated values. Thereafter, the trained classifier was used to carry out the cascade feature screening. Once the feature had passed through all of the strong classifier screening, it was determined to be the face of the region. (b) Using the integral image on the Haar-like features to accelerate the evaluation: characteristic values were calculated using a feature rectangle, which was related only to the integral image of the feature endpoint, regardless of the image coordinate's value. Therefore, regardless of the scale of the feature rectangle, the time taken to perform the calculation of the characteristic values was constant because it only involved addition and subtraction calculations. For this reason, the introduction of the integral image greatly improved the detection speed. (c) The AdaBoost training algorithm was used to distinguish faces and non-faces with a strong classifier. (d) The screening style cascade to strong classifier cascade together, to increase the accuracy. The acquisition of the expert's acupoint data could be used along with the mouse to complete the acupoint data acquisition, and to collect values of acupoint location. Consistency information fusion algorithm of acupuncture experts' acupoint location The algorithm was based on the consistency matrix pointwise fusion method between the measured values,13-18 following these steps: 1. A weight k i was given to each acupuncture expert based on their job title and familiarity with the business. For example, first a basic score k1i was deter-
Using a high-efficiency rectangle features and integral image to extract the underlying characteristics The software used a Haar cascade classifier with an Adaboost algorithm,13 which trained a strong cascading classifier, and at the bottom level of the feature extraction, used a highly efficient method which focused on JTCM | www. journaltcm. com
112
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
mined based on job title: professor, k11 = 0.9-1.0; associate professor, k12 = 0.8-0.9; lecturer, k13 = 0.7-0.8. Then, a score based on clinical practice was given according to each expert's clinical practice ability, producing k 2i = 0.00-0.09, k i = k1i + k 2i . 2. The system error correction coefficient r i was calculated based on the difference in distance between each acupuncture expert's acupoint location data point's mean value and the expectations of all acupuncture experts' mean values, to determine each expert's acupoint location system errors. (a) First, the mean value of an acupoint Aˉix , Aˉiy was calculated for each acupuncture expert (i represents the i-th acupuncture expert); (b) Then, the expected value Aˉx , Aˉy of the mean value of a plurality of acupuncture experts was calculated; (c) Finally, using the distance Hˉi between the mean value of the measurements of each acupuncture expert and the expected value of the measurement, and the average distance Hˉ of a plurality of acupuncture experts, the systematic error correction coefficient was calculatHˉ (where again i represents the i-th acupunced: r i = Hi ture expert). 3. An acupoint on the body's surface is a circular area with a diameter of about 5-8 mm.19,20 Therefore, the surface area of the acupoint at each expert location, using the intersection set of the measurement results of the same measurement target's acupoint, can be used to calculate the consistency of data to get a weight for each measured value. (a) The consistency support weights were calculated using the i-th expert measured value relative to the j-th Ci ⋂ C j expert measured value, where d ij = , ( C i ⋂ C j is C the size of the intersecting area that is produced by measuring the point area of the i-th and j-th acupuncture experts. C is the circular area of the acupoint on the body's surface; numerically, C = π × R2 = 3.14 × 3.252 = 33.17 mm2. i, j are the two different acupuncture experts, and d ij is the i-th row and j-th column value of the consistency support matrix D ij ). (b) Summing the values of the D ij line obtained all experts for the i-th expert's consistency support degree l i of the measured values. This method was used to find all of the measured values of an acupuncture expert, which were affected by other effects of acupuncture consistency support. (c) The consistency support weights of the measured A
b
a c
C
B a
b c
value were calculated based on the consistency support degree of the value measured by acupuncture experts. 4. To calculate multiple expert fusion values of an acupoint, the systematic error was corrected to obtain consistency support weights for the measured values of the same acupoints that were located by different experts. (a) The fusion value of the measurement result of the acupuncture expert and the fusion value of the measured value of the error were calculated based on the weight of each expert k i , the system error correction factor r i , and the consistency support weight of measured value m i , where k i r i m i is the k i × r i × m i n
value after normalization. X ' =∑k i r i m i X i ( X i is the i =1
actual measured value of the i-th expert in acupoint points), where X ' was the fusion result; n
δ' x = ∑k i r i m i δ ix ( δ ix is the actual measured error i =1
of the i-th expert in acupoint points), where δ' x is the n
fusion boundary; Y ' =∑k i r i m iY i ( Y i is the actual i =1
measured value of the i-th expert in acupoint points); n
δ' y = ∑k i r i m i δ iy ( δ iy is the actual measured error of i =1
the i-th expert in acupoint points), where δ' y is the fusion boundary. Figure 2 shows the combination methods of the information source result range of three expert's acupoint locations. In addition to the combinations shown in Figure 2, a situation that occurred with zero probability where the three points a, b and c coincided on the same coordinate point. The consistency information fusion algorithm was used to acquire the optimal combination of different expert's acupoint location information. After algorithm processing, the amount of information fusion, i.e., each of expert information could be fusioned, confirmed each other, was used to get the best precision and accuracy of the acupoint values. The acupoint data were digitized, objectified, standardized and normalized, and built the acupoint database. Visualization learning interface of human-computer interaction techniques The system provided a visual learning environment for practitioners in the computer manipulation of virtual environments,21 using the best location data from acupuncture expert's to highlight practicing, self-testing and learning.
D a
b
a
b
c
c
Figure 2 Combination methods of the information source result range of three expert's acupoint location A: seperation type; B: seperation and intersection type; C: intersection and seperation type; D: each of the two intersceting type. a, b, c respectively meant result range of three expert's acupoint location. JTCM | www. journaltcm. com
113
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
lated Hˉi1 = 2.1818, Hˉi2 = 0.6250, Hˉi3 = 3.3279, Hˉi4 = 2.3773, Hˉi5 = 1.3099, Hˉi6 = 2.662, based on the difference in distance between the average value of the measured data from each expert's acupoint location and the average value of all acupuncture experts expectation, and used the average value of distance of a plurality of acupuncture experts to obtain the system error correction factor: r1 = 0.9231, r 2 = 3.2216, r3 = 0.6052, r i = 0.8486, r 5 = 1.5376, and r 6 = 0.8887. Acupoints on the body surface have a diameter of about 5–8 mm (we take 6.5 mm as the average diameter). Table 2 shows the Chengqi (ST 1) acupoint location of the six experts based on the intersection of the measurement results between the body surface area of each expert acupoint location point and the other experts on the same measurement target. From this we obtained matrix D ij . é1.0000 0.4706 0.0705 0.2617 0.7360 0.8101ù ê0.4706 1.0000 0.4782 0.6259 0.6327 0.4548ú Di j = ê0.0705 0.4782 1.0000 0.4355 0.2110 0.0821ú ê0.2617 0.6259 0.4355 1.0000 0.3266 0.1969ú ê0.7360 0.6327 0.2110 0.3266 1.0000 0.8036ú ë0.8101 0.4548 0.0821 0.1969 0.8036 1.0000û (1) l i is the sum of row i in Eq.(1):
The learning assessment method was based on the coordinate value of the acupoint in the database. The practitioner's acupoint location ability and score were determined based on the size of the intersecting area between the expert's coordinate acupoint values in the database and the coordinate point values that were entered by the practitioners in the visual learning interface. The system used a simple block design and made the practitioners' interface clear and easy to operate. By comparing the score between the expert and the practitioner, the practitioner could obtain a more intuitive and objective understanding of location gaps.
RESULTS Because of their differences in accuracy and acupoint location ability, every expert had a unique system error. Under these conditions, the measure of every expert had likely his location problem. This would give the result misdirection. Thus, a modification factor was used to correct all instances of system errors to obtain credible and precise results. Six acupuncture experts with rich acupuncture experience from Acupuncture and Tuina College, Shanghai University of TCM, were selected for Chengqi (ST 1) acupoint location. Their average age was 40 years old. Three were male, three were female, three were professors and three were associate professors. Each expert located the acupoint three times, and their mean values and standard deviations are shown in Table 1. According to acupuncture experts' acupoint location, each acupuncture expert's weight k i was k1 = 0.85, k 2 = 0.85, k3 = 0.80, k 4 = 0.90, k 5 = 0.90, and k 6 = 0.90. From the coordinate data in Table 1, we obtained the expectations Aˉx = ﹣39.5750, Aˉy = 9.0067. We calcu-
n
l i =∑d ij , (2) j =1
where n is the number of experts. Thus, l i1 = 3.3485, l i2 = 3.6622, l i3 = 2.2773, l i4 = 2.8467, l i5 = 3.7089, and l i6 = 3.3474. The weight m i was obtained from the normalization of l i , and stands for quantification of consistency. m i1 = 0.1745, m i2 = 0.1908, m i3 = 0.1187, m i4 = 0.1483, m i5 = 0.1933, and m i6 = 0.1744. Finally, we obtained: k1 × r1 × m1 = 0.1300, k 2 × r 2 × m 2 = 0.5200, k3 × r3 × m3 = 0.0580, k 4 × r 4 × m 4 = 0.1100, k 5 r 5 m 5 = 0.2600, and
Table 1 Mean values and standard deviations of six experts on the Chengqi (ST 1) acupoint location Numerical
Chengqi X mean value
Chengqi Y mean value
Chengqi X standard deviation
Chengqi Y standard deviation
Expert 1
﹣40.3976
11.0274
0.0431
0.0867
Expert 2
﹣39.4876
8.3877
0.3305
0.7189
Expert 3
﹣38.0026
6.0736
0.0316
0.0480
Expert 4
﹣40.8407
7.0031
0.1861
0.0358
Expert 5
﹣39.2624
10.2787
0.0852
0.2531
Expert 6
﹣39.4528
11.2695
0.0451
0.1762
Table 2 Size of the Chengqi (ST 1) acupoint location intersecting circle areas across experts Intersecting circle area
Expert 1
Expert 2
Expert 3
Expert 4
Expert 5
Expert 6
Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Expert 6
33.1663 15.6084 2.3394 8.6812 24.4089 26.8672
15.6084 33.1663 15.8599 20.7600 20.9840 15.0833
2.3394 15.8599 33.1663 14.4436 6.9986 2.7218
8.6812 20.7600 14.4436 33.1663 10.8322 6.5301
24.4089 20.9840 6.9986 10.8322 33.1663 26.6517
26.8672 15.0833 2.7218 6.5301 26.6517 33.1663
114
February 15, 2015 | Volume 35 | Issue 1 |
JTCM | www. journaltcm. com
Yang XM et al. / Methodology
k 6 × r 6 × m 6 = 0.1400. k i r i m is the normalized value of k i × r i × m i , k1 r1m1 = 0.1067, k 2 r 2 m 2 = 0.4269, k3 r3 m3 = 0.0476, k 4 r 4 m 4 = 0.0928, k 5 r 5 m 5 = 0.2110, and k 6 r 6 m 6 = 0.1149.
components (time, attention, credibility, expectation), specific non-needling components (psychological: history, diagnosis, education; physiological: palpation, moxibustion), needling components (location, insertion depth, stimulation, needle size and number). This lack of understanding of what was and what was not an acupuncture acupoint impacts our ability to interpret the results of clinical trials that compare needling at acupuncture and nonacupuncture (i.e., sham) acupoints. It is now the important responsibility of acupuncture researchers to face these results squarely and move the field forward. Any variability in acupuncture treatment results which occurred because of unreliable and/or inaccurate acupoint location methods might eventually be reflected in both clinical practice and research trials. Accurate and precise acupoint location will enable researchers to uncover the mystery of the two provocative paradoxes, creating the preconditions and foundation for their solution. Acupuncture treatment of disease has a long history, and accurate and precise acupoint location is its foundation. Moreover, the lack of proper location is likely to make what has emerged as two provocative paradoxes a permanent mystery. Australian scholars have developed two contemporary methods that are better than traditional acupuncture in terms of precision and accuracy, but in terms of participants' perceived degree of comfort, the two contemporary methods are less comfortable than the two traditional methods. According to the basic theory of the Chinese art of face structure theory, an established acupoint location method, not standardizing the target face images may lead to lower rates of acupoint location in terms of precision and accuracy. Moreover, because of their special nature, most facial acupoint locations were unsuitable for the "bone standard" or "finger cun measurement" methods, so only the "anatomical landmark" method could be used. Therefore, in this paper, each correct acupoint location data point was obtained by acupuncture experts, and a variable precision method was proposed to improve weight computing. This method introduced weights for each expert, system error, and consistency. According to the method, weight of reliable data had been increased, so a higher reliability value could be got. Compared with other constant precision results, we could obtain a precise and accurate acupoint location value. Accurate and precise acupoint location for acupuncture treatments should be studied (a) "top down" as multi-component whole-system interventions, and (b) "bottom up" as mechanistic studies that focus on understanding how individual treatment components interact and translate into clinical and physiological outcomes.23 Such a strategy, incorporating considerations of efficacy, effectiveness and qualitative measurement, would strengthen the evidence base for such complex interventions as acupuncture. The alternative is to research efforts toward understanding the fundamental
(3) n
Y ' = ∑k i r i m iY i (4) i =1 n
δ x = ∑k i r i m i δ ix (5) '
i =1 n
δ y = ∑k i r i m i δ iy (6) '
i =1
X i , Y i is the actual measured value of the i-th expert acupuncture location acupoint, while δ ix , δ iy is the actual measured error of the i-th expert acupuncture location acupoint. The fusion "pointwise variable precision result" was calculated as: X' = ﹣ 39.5845, δ x' = 0.1865; Y' = 9.1599, δ y ' = 0.3954. This method was based on the original data inconsistency, and could not be achieved using the characteristics of the general weights calculation. Through correcting the system errors of various experts' acupoint location, reasonably quantifying the consistency of the locations, and obtaining fusion results of pointwise variable precision, measurements could more effectively be combined from different experts' acupoint location to increase the efficiency of measurement information. Patient welfare, clinical efficacy, and valid research results depend on accurate and precise acupoint location. A few years ago, the Australian scholar Mark Aird proposed that 22 improving the precision of acupoint location techniques could be seen as a three step process starting with measuring the required precision, then developing a more precise method, and finally identifying a procedure to measure whether an acupoint had been correctly located. Here, we have completed the first two steps.
DISCUSSION The accuracy and precision of acupoint location methods have important implications for research. Treatment outcome in clinical trials might be significantly altered if an inaccurate method of acupoint location was used. The task at hand was to instruct practitioners in the most precise and accurate method, rather than sacrificing experimental validity and clinical effectiveness for adherence to other methods. While the most likely consequences of imprecise acupoints are variability in treatment effects, there is also the potential for harm to patients and research participants if acupoints were not precisely located. Acupoint location descriptions and methods are vital considerations in research and practice. The field of acupuncture research has matured considerably, yet two provocative paradoxes have emerged.23 Acupuncture treatment components were made of non-specific JTCM | www. journaltcm. com
115
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
theories and practices of acupuncture, including the ability to precisely locate acupoints. Accurate and precise acupoint location has great significance for improving the curative effects of acupuncture. We collected information about acupuncture experts' acupoint locations on the face, and used an information fusion algorithm to obtain the most accurate and precise acupoint location values. Meanwhile, we developed a virtual application system to construct facial acupoint locations, to provide a platform for the learning of facial acupoint location, and to obtain the most accurate and precise acupoint location values on the face.
10
REFERENCES
15
1 2 3
4
5 6
7
8
9
11
12
13 14
Lao LX. Acupuncture situation and to explore clinical research. Zhen Ci Yan Jiu 2008; 33(1): 53-58. Liang FR, Ren YL, Tang Y. Acupuncture clinical evaluation. Zhen Ci Yan Jiu 2008; 33(1): 62-64. Yang XM, Xu LY, Zhong F, et al. Data mining-based detection of acupuncture treatment on juvenile myopia. Zhong Yi Za Zhi 2012; 32(3): 372-376. Liu WZ, Duan GJ. Acupuncture on the visual system and its mechanism of action research progress. Zhong Yi Yan Ke Za Zhi 1999; 5(3): 180. Chen XN. Chinese acupuncture and moxibustion. Beijing: Foreign Languages Publishing House, 2002: 60. Coyle MG, Mark A, Cobbin DM, Zaslawski C. The cun measurement system: an investigation into its suitability in current practice. Acupuncture Med 2000; 18(1): 10-14. Eckman P. In the footsteps of the yellow emperor: tracing the history of traditional acupuncture. San Francisco: Cypress Book Company, 1996: 66-70. Mark A, Deirder M, Carole R, Carole R. A study of the relative precision of acupoint location methods. J Altern Complement Med 2002; 5(8): 635-642. Aird M, Coyle ME, Cobbin DM, Zaslawski CJ. A study of the reliability of two methods of location acupuncture points. Acupunc Med 2000; 18: 15-21.
JTCM | www. journaltcm. com
16
17
18
19 20
21 22
23
116
Zhao Y, Zhang T, Lian QS. Infrared facial image Bell's palsy acupuncture location algorithm. Zhong Guo Ji Suan Ji Xi Tong Za Zhi 2012; 33(7): 1613-1619. Yan ZG. English-Chinese layer anatomical atlas of acupoints of the human body. Shanghai: University of Traditional Chinese Medicine Publishing House, 2005: 72. Wei LH, Wang W, Meng XM, et al. Principle in selecting acupoint and the prescription rule in acupuncture and moxibustion therapy. Tianjin Zhong Yi Yao Da Xue Xue Bao 2008; 27(2): 59-60. Zhao N. Face detection based on AdaBoost. Beijing: University of Beijing, 2005: 16-28. Xu LY. Source selection and dynamic feature extraction sequence level information fusion model and algorithm research. Shenyang: Northeastern University, 2001: 1-11. Zhon F, Liu N, Liu Y, Xu LY. New SST correction method from multi-satellite based on the coefficient of variation. Shanghai Da Xue Xue Bao 2011; 15(5): 463-466. Xu LY, Zhong F, Zhang PN. Conflict analysis of multi-source SST distribution. Shu Zi Feng Xi Yu Mo Xin Za Zhi 2012; 9(2): 338-342. Shi SX. Study on soft fusion strategy of multi-channel incertitude information in digital ocean. Shenyang: Northeastern University, 2005: 31-59. Wang J. Multi satellite monitoring SST data variable precision / reliability assessment studies. Shanghai: Shanghai University, 2011: 26-60. Wang XB. Study abroad meridian problems. Beijing: People's Health Publishing House, 1984: 9-15. Xiao YJ, Xiao P. Discussion acupoints on body surface area. Shandong Zhong Yi Yao Da Xue Xue Bao 1999; 23(1): 11-18. Zhu ZX. Engineering psychology. Shanghai: East China Normal University Publishing House, 1990: 10-15. Mark A. Variability in the precision of acupoint location methods. Sydney: University of Technology Sydney, 2005: 15-19. Langevin HM, Wayne PM, Macpherson H, et al. Paradoxes in acupuncture research: strategies for moving forward. Evid Based Complement Alternat Med 2011; 2011: 180805.
February 15, 2015 | Volume 35 | Issue 1 |
J Tradit Chin Med 2015 February 15; 35(1): 110-116 ISSN 0255-2922 © 2015 JTCM. All rights reserved.
METHODOLOGY TOPIC
A precise and accurate acupoint location obtained on the face using consistency matrix pointwise fusion method
Yang Xuming, Ye Yijun, Xia Yong, Wei Xuanzhong, Wang Zheyu, Ni Hongmei, Zhu Ying, Xu Lingyu aa Yang Xuming, Ye Yijun, Xia Yong, Acupuncture and Tuina College, Shanghai University of Traditional Chinese Medicine, Shanghai 201203, China Wei Xuanzhong, Wang Zheyu, School of Electronic Information and Electrical Engineering, Shanghai Jiaotong University, Shanghai 200240, China Ni Hongmei, Basic Medical College, Shanghai University of Traditional Chinese Medicine, Shanghai 201203, China Zhu Ying, Ophthalmology Department, Shanghai of Traditional Chinese Medicine Hospital, Shanghai 200071, China Xu Lingyu, College of Computer Engineering and Science, Shanghai University, Shanghai 200072, China Supported by the Key Program of State Administration of Traditional Chinese Medicine of China: the Science of Acupuncture and Moxibustion (No. ZYSNXD-CC-ZDXK-07) Correspondence to: Prof. Yang Xuming, Acupuncture and Tuina College, Shanghai University of Traditional Chinese Medicine, Shanghai 201203, China. [email protected] Telephone: +86-21-51322434; +86-18916005529 Accepted: February 27, 2014
the general weight calculation. First, we corrected each expert of acupoint location system error itself, to obtain a rational quantification for each expert of acupuncture and moxibustion acupoint location consistent support degree, to obtain pointwise variable precision fusion results, to put every expert's acupuncture acupoint location fusion error enhanced to pointwise variable precision. Then, we more effectively used the measured characteristics of different acupuncture expert's acupoint location, to improve the measurement information utilization efficiency and acupuncture acupoint location precision and accuracy.
Abstract
© 2015 JTCM. All rights reserved.
OBJECTIVE: To develop a more precise and accurate method, and identified a procedure to measure whether an acupoint had been correctly located.
Key words: Information fusion; Face recognition; Facial acupoint; Acupoint location
CONCLUSION: Based on using the consistency matrix pointwise fusion method on the acupuncture experts' acupoint location values, each expert's acupoint location information could be calculated, and the most precise and accurate values of each expert's acupoint location could be obtained.
INTRODUCTION
METHODS: On the face, we used an acupoint location from different acupuncture experts and obtained the most precise and accurate values of acupoint location based on the consistency information fusion algorithm, through a virtual simulation of the facial orientation coordinate system.
The effectiveness of acupuncture and its safe and scientific characteristics have been fully affirmed by the World Health Organization (WHO) and the National Institutes of Health (NIH), and acupuncture has become an important parts of the world's medicine.1,2 Many facial diseases,3 such as juvenile myopia, xerophthalmia, rhinitis and facial paralysis have shown significant treatment effects following acupuncture.4 Correct acupoint location is the first step of acupuncture therapy
RESULTS: Because of inconsistencies in each acupuncture expert's original data, the system error could not be modified using the characteristics of JTCM | www. journaltcm. com
110
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
and acupuncture treatment for facial diseases, because the facial capillaries and nerves are extremely rich, which is the key to the safety and curative effects of acupuncture and moxibustion. Three kinds of traditional acupoint location methods consist of basing location on anatomical landmarks of the surface of the body,5 measuring the distance between each part in Chinese anatomical inches (cun), and using the patient's fingers in proportion to a cun ruler. The cun measurement system is an essential component of traditional acupoint location methods used in acupuncture, dating back to 2900 BC, when pieces of jade resembling rulers were used for measurement.6,7 In 2002, an Australian scholar performed experiments8 on 72 participants selected from undergraduates of the Bachelor of Health Sciences in Acupuncture course, who repeatedly attempted to locate a fictitious acupoint (FP), with the resulting coordinates recorded. Significant differences were found between the two traditional methods (directional and proportional) and two contemporary methods (elastic and ruler). The 95% confidence ellipse for the ruler method had an area of 2.9 cm2, and that for the directional method had an area of 12.7 cm2. However, in terms of ease of use, most of the ratings for the traditional methods were 'comfortable', while almost half of the ratings for the two variant methods were 'uncomfortable'. Unfortunately, the two more precise contemporary methods were generally not well received; the two traditional methods were not with high precision, but which could comfortably provide with percived appearance to patients.9 In 2012, following the facial structure theory,10 Chinese scholars partitioned the facial features, using the Minimum Eigenvalue operator to detect the corner points of the features and the Log operator to detect the edges of the features; the comprehensive application of corner and edge information was used to localize each facial feature position. Finally, using the facial feature location as a reference coordinate, the facial acupoint was located, and according to the difference between the left and right point temperature of the face, the Baer facial acupuncture acupoint therapy was automatically selected. However, an abnormal target facial image could lead to a low accuracy rate of acupoint location. Because of the special nature of the face, in the metheds of "bone standard","finger cun measurement"and "anatomical landmark", the vast majority of acupoint used the "anatomical landmark" acupoint location method. An example is Chenqi (ST 1) location: on the face, directly below the pupil, in the depression of the infraorbital foramen.11 Here, we only discuss cases of acupoint location that used the "anatomical landmark" method. In short, the accuracy and precision of acupoint location methods have important implications for research. Treatment outcomes in clinical trials could be significantly altered if an inaccurate method of acupoint locaJTCM | www. journaltcm. com
tion was used. Practitioners must be able to locate on an acupoint with sufficient precision and accuracy to avoid stimulating nearly acupoints, since they might have effects other than those desired. At the same time, practitioners rely on different acupoints that have different physiological effects. Patient welfare, clinical efficacy, and valid research results depend on accurate and precise acupoint location. Moreover, different acupuncture prescriptions have the specific effect of treating different diseases.12 Therefore, we implemented an accurate and precise acupoint location learning method on the face, which was characterized by: establishing a facial orientation coordinate system with a virtual simulation, collecting acupoint location information from acupuncture experts, and using the consistency fusion algorithm for acupoint location, to get the most accurate and precise acupoint location values. The specific steps were as follows: (a) On the face, locating the acupoint and labelling it, then collecting front face photos and importing them into the computer; (b) In the computer's virtual environment, using local facial feature analysis and graphics and a neural recognition algorithm to determine the human facial organs and features, to build the coordinate origin and establish the coordinate system; (c) With multiple sources of information in time and spatial coordinates and the advantage of complementary and redundant information in common or joint operation, using the consistency information fusion algorithm, by fusing acupoint values from different acupuncture experts, to obtain precise and accurate coordinate values; (d) Establishing an acupoint database and constructing an interactive visualization learning interface, which lets the acupuncture practitioners who used the expert location values in the computer manipulation of the virtual learning environment establish acupoint location and prepare acupuncture prescription.
DATA ACQUISITION METHOD OF FACIAL ACUPOINTS Following the principles of Traditional Chinese Medicine (TCM), an acupoint location method was determined that suited the "anatomical landmark" method, collecting different acupuncture experts' acupoint location values to produce the best accuracy and precision values of each acupoint obtained via the consistency information fusion algorithm, providing a platform of facial acupoint location learning. Facial recognition technology to create a virtual simulation location system The specific methods were as follows. In a computer virtual environment, partial facial feature analysis and a graph and nerve recognition algorithm were used to determine the characteristics of the body's facial organs and features. Next, the origin of the coordinates was determined and the coordinate system was established. 111
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
Locating and identifying the acupoint on face Acquiring the front face photos Using the computer face recognition technology, to determine the human face organs and characteristic of the region, establishing coordinate system Getting acupuncture expert's acupoint location values By the technology of information fusion, calculating each acupuncture point "the best precise value" Establishing expert knowledge base of acupuncture acupoint Constructing digital learning interface of human-computer interaction Figure 1 One kind of precise facial acupoint location learning method flowchart
(a) Facial recognition procedure: to establish the facial image archives, facial images of each person were taken using a camera or were inputted from other photos to form the face image files, and were then transformed into Faceprint code that was stored in the archive base. A current Faceprint was generated from human face images taken by the camera, or from inputted pictures. The current Faceprint was compared with the Faceprint that was stored in the archive base. The Faceprint approach was based on the essential face characteristics and the facial shape. It could ignore light, skin tone, facial hair, hairstyle, glasses, facial expressions and posture change with strong reliability, so that it could accurately recognize the person from the pictures. The face recognition process, using common image processing software, could be completed automatically and continuously, in real time. (b) Establishment of the coordinate system: the algorithm was used to identify the face, and then the origin of a coordinate axis for the midpoint of the two eyes was identified, using the straight line between pupils for the X-axis. Because it was impossible to ensure that participants sat exactly in the same place, the X-axis of each photo coordinate system was not at the same level. Thus, the facial data were normalized to guarantee that they were comparable. In each photo, the person's own facial properties were used as a reference to establish the coordinate system. In practice, intended to the interpupillary distance (PD) 1/80 of a unit, to solve the problem that everyone's face is a different size.
rectangular features and the integral image. The algorithm was: (a) Using Haar-like features for testing: haar-like features: a region was characterized by means of facial detection under a sub-window. The pixels of the region were based on certain formula-calculated values. Thereafter, the trained classifier was used to carry out the cascade feature screening. Once the feature had passed through all of the strong classifier screening, it was determined to be the face of the region. (b) Using the integral image on the Haar-like features to accelerate the evaluation: characteristic values were calculated using a feature rectangle, which was related only to the integral image of the feature endpoint, regardless of the image coordinate's value. Therefore, regardless of the scale of the feature rectangle, the time taken to perform the calculation of the characteristic values was constant because it only involved addition and subtraction calculations. For this reason, the introduction of the integral image greatly improved the detection speed. (c) The AdaBoost training algorithm was used to distinguish faces and non-faces with a strong classifier. (d) The screening style cascade to strong classifier cascade together, to increase the accuracy. The acquisition of the expert's acupoint data could be used along with the mouse to complete the acupoint data acquisition, and to collect values of acupoint location. Consistency information fusion algorithm of acupuncture experts' acupoint location The algorithm was based on the consistency matrix pointwise fusion method between the measured values,13-18 following these steps: 1. A weight k i was given to each acupuncture expert based on their job title and familiarity with the business. For example, first a basic score k1i was deter-
Using a high-efficiency rectangle features and integral image to extract the underlying characteristics The software used a Haar cascade classifier with an Adaboost algorithm,13 which trained a strong cascading classifier, and at the bottom level of the feature extraction, used a highly efficient method which focused on JTCM | www. journaltcm. com
112
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
mined based on job title: professor, k11 = 0.9-1.0; associate professor, k12 = 0.8-0.9; lecturer, k13 = 0.7-0.8. Then, a score based on clinical practice was given according to each expert's clinical practice ability, producing k 2i = 0.00-0.09, k i = k1i + k 2i . 2. The system error correction coefficient r i was calculated based on the difference in distance between each acupuncture expert's acupoint location data point's mean value and the expectations of all acupuncture experts' mean values, to determine each expert's acupoint location system errors. (a) First, the mean value of an acupoint Aˉix , Aˉiy was calculated for each acupuncture expert (i represents the i-th acupuncture expert); (b) Then, the expected value Aˉx , Aˉy of the mean value of a plurality of acupuncture experts was calculated; (c) Finally, using the distance Hˉi between the mean value of the measurements of each acupuncture expert and the expected value of the measurement, and the average distance Hˉ of a plurality of acupuncture experts, the systematic error correction coefficient was calculatHˉ (where again i represents the i-th acupunced: r i = Hi ture expert). 3. An acupoint on the body's surface is a circular area with a diameter of about 5-8 mm.19,20 Therefore, the surface area of the acupoint at each expert location, using the intersection set of the measurement results of the same measurement target's acupoint, can be used to calculate the consistency of data to get a weight for each measured value. (a) The consistency support weights were calculated using the i-th expert measured value relative to the j-th Ci ⋂ C j expert measured value, where d ij = , ( C i ⋂ C j is C the size of the intersecting area that is produced by measuring the point area of the i-th and j-th acupuncture experts. C is the circular area of the acupoint on the body's surface; numerically, C = π × R2 = 3.14 × 3.252 = 33.17 mm2. i, j are the two different acupuncture experts, and d ij is the i-th row and j-th column value of the consistency support matrix D ij ). (b) Summing the values of the D ij line obtained all experts for the i-th expert's consistency support degree l i of the measured values. This method was used to find all of the measured values of an acupuncture expert, which were affected by other effects of acupuncture consistency support. (c) The consistency support weights of the measured A
b
a c
C
B a
b c
value were calculated based on the consistency support degree of the value measured by acupuncture experts. 4. To calculate multiple expert fusion values of an acupoint, the systematic error was corrected to obtain consistency support weights for the measured values of the same acupoints that were located by different experts. (a) The fusion value of the measurement result of the acupuncture expert and the fusion value of the measured value of the error were calculated based on the weight of each expert k i , the system error correction factor r i , and the consistency support weight of measured value m i , where k i r i m i is the k i × r i × m i n
value after normalization. X ' =∑k i r i m i X i ( X i is the i =1
actual measured value of the i-th expert in acupoint points), where X ' was the fusion result; n
δ' x = ∑k i r i m i δ ix ( δ ix is the actual measured error i =1
of the i-th expert in acupoint points), where δ' x is the n
fusion boundary; Y ' =∑k i r i m iY i ( Y i is the actual i =1
measured value of the i-th expert in acupoint points); n
δ' y = ∑k i r i m i δ iy ( δ iy is the actual measured error of i =1
the i-th expert in acupoint points), where δ' y is the fusion boundary. Figure 2 shows the combination methods of the information source result range of three expert's acupoint locations. In addition to the combinations shown in Figure 2, a situation that occurred with zero probability where the three points a, b and c coincided on the same coordinate point. The consistency information fusion algorithm was used to acquire the optimal combination of different expert's acupoint location information. After algorithm processing, the amount of information fusion, i.e., each of expert information could be fusioned, confirmed each other, was used to get the best precision and accuracy of the acupoint values. The acupoint data were digitized, objectified, standardized and normalized, and built the acupoint database. Visualization learning interface of human-computer interaction techniques The system provided a visual learning environment for practitioners in the computer manipulation of virtual environments,21 using the best location data from acupuncture expert's to highlight practicing, self-testing and learning.
D a
b
a
b
c
c
Figure 2 Combination methods of the information source result range of three expert's acupoint location A: seperation type; B: seperation and intersection type; C: intersection and seperation type; D: each of the two intersceting type. a, b, c respectively meant result range of three expert's acupoint location. JTCM | www. journaltcm. com
113
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
lated Hˉi1 = 2.1818, Hˉi2 = 0.6250, Hˉi3 = 3.3279, Hˉi4 = 2.3773, Hˉi5 = 1.3099, Hˉi6 = 2.662, based on the difference in distance between the average value of the measured data from each expert's acupoint location and the average value of all acupuncture experts expectation, and used the average value of distance of a plurality of acupuncture experts to obtain the system error correction factor: r1 = 0.9231, r 2 = 3.2216, r3 = 0.6052, r i = 0.8486, r 5 = 1.5376, and r 6 = 0.8887. Acupoints on the body surface have a diameter of about 5–8 mm (we take 6.5 mm as the average diameter). Table 2 shows the Chengqi (ST 1) acupoint location of the six experts based on the intersection of the measurement results between the body surface area of each expert acupoint location point and the other experts on the same measurement target. From this we obtained matrix D ij . é1.0000 0.4706 0.0705 0.2617 0.7360 0.8101ù ê0.4706 1.0000 0.4782 0.6259 0.6327 0.4548ú Di j = ê0.0705 0.4782 1.0000 0.4355 0.2110 0.0821ú ê0.2617 0.6259 0.4355 1.0000 0.3266 0.1969ú ê0.7360 0.6327 0.2110 0.3266 1.0000 0.8036ú ë0.8101 0.4548 0.0821 0.1969 0.8036 1.0000û (1) l i is the sum of row i in Eq.(1):
The learning assessment method was based on the coordinate value of the acupoint in the database. The practitioner's acupoint location ability and score were determined based on the size of the intersecting area between the expert's coordinate acupoint values in the database and the coordinate point values that were entered by the practitioners in the visual learning interface. The system used a simple block design and made the practitioners' interface clear and easy to operate. By comparing the score between the expert and the practitioner, the practitioner could obtain a more intuitive and objective understanding of location gaps.
RESULTS Because of their differences in accuracy and acupoint location ability, every expert had a unique system error. Under these conditions, the measure of every expert had likely his location problem. This would give the result misdirection. Thus, a modification factor was used to correct all instances of system errors to obtain credible and precise results. Six acupuncture experts with rich acupuncture experience from Acupuncture and Tuina College, Shanghai University of TCM, were selected for Chengqi (ST 1) acupoint location. Their average age was 40 years old. Three were male, three were female, three were professors and three were associate professors. Each expert located the acupoint three times, and their mean values and standard deviations are shown in Table 1. According to acupuncture experts' acupoint location, each acupuncture expert's weight k i was k1 = 0.85, k 2 = 0.85, k3 = 0.80, k 4 = 0.90, k 5 = 0.90, and k 6 = 0.90. From the coordinate data in Table 1, we obtained the expectations Aˉx = ﹣39.5750, Aˉy = 9.0067. We calcu-
n
l i =∑d ij , (2) j =1
where n is the number of experts. Thus, l i1 = 3.3485, l i2 = 3.6622, l i3 = 2.2773, l i4 = 2.8467, l i5 = 3.7089, and l i6 = 3.3474. The weight m i was obtained from the normalization of l i , and stands for quantification of consistency. m i1 = 0.1745, m i2 = 0.1908, m i3 = 0.1187, m i4 = 0.1483, m i5 = 0.1933, and m i6 = 0.1744. Finally, we obtained: k1 × r1 × m1 = 0.1300, k 2 × r 2 × m 2 = 0.5200, k3 × r3 × m3 = 0.0580, k 4 × r 4 × m 4 = 0.1100, k 5 r 5 m 5 = 0.2600, and
Table 1 Mean values and standard deviations of six experts on the Chengqi (ST 1) acupoint location Numerical
Chengqi X mean value
Chengqi Y mean value
Chengqi X standard deviation
Chengqi Y standard deviation
Expert 1
﹣40.3976
11.0274
0.0431
0.0867
Expert 2
﹣39.4876
8.3877
0.3305
0.7189
Expert 3
﹣38.0026
6.0736
0.0316
0.0480
Expert 4
﹣40.8407
7.0031
0.1861
0.0358
Expert 5
﹣39.2624
10.2787
0.0852
0.2531
Expert 6
﹣39.4528
11.2695
0.0451
0.1762
Table 2 Size of the Chengqi (ST 1) acupoint location intersecting circle areas across experts Intersecting circle area
Expert 1
Expert 2
Expert 3
Expert 4
Expert 5
Expert 6
Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Expert 6
33.1663 15.6084 2.3394 8.6812 24.4089 26.8672
15.6084 33.1663 15.8599 20.7600 20.9840 15.0833
2.3394 15.8599 33.1663 14.4436 6.9986 2.7218
8.6812 20.7600 14.4436 33.1663 10.8322 6.5301
24.4089 20.9840 6.9986 10.8322 33.1663 26.6517
26.8672 15.0833 2.7218 6.5301 26.6517 33.1663
114
February 15, 2015 | Volume 35 | Issue 1 |
JTCM | www. journaltcm. com
Yang XM et al. / Methodology
k 6 × r 6 × m 6 = 0.1400. k i r i m is the normalized value of k i × r i × m i , k1 r1m1 = 0.1067, k 2 r 2 m 2 = 0.4269, k3 r3 m3 = 0.0476, k 4 r 4 m 4 = 0.0928, k 5 r 5 m 5 = 0.2110, and k 6 r 6 m 6 = 0.1149.
components (time, attention, credibility, expectation), specific non-needling components (psychological: history, diagnosis, education; physiological: palpation, moxibustion), needling components (location, insertion depth, stimulation, needle size and number). This lack of understanding of what was and what was not an acupuncture acupoint impacts our ability to interpret the results of clinical trials that compare needling at acupuncture and nonacupuncture (i.e., sham) acupoints. It is now the important responsibility of acupuncture researchers to face these results squarely and move the field forward. Any variability in acupuncture treatment results which occurred because of unreliable and/or inaccurate acupoint location methods might eventually be reflected in both clinical practice and research trials. Accurate and precise acupoint location will enable researchers to uncover the mystery of the two provocative paradoxes, creating the preconditions and foundation for their solution. Acupuncture treatment of disease has a long history, and accurate and precise acupoint location is its foundation. Moreover, the lack of proper location is likely to make what has emerged as two provocative paradoxes a permanent mystery. Australian scholars have developed two contemporary methods that are better than traditional acupuncture in terms of precision and accuracy, but in terms of participants' perceived degree of comfort, the two contemporary methods are less comfortable than the two traditional methods. According to the basic theory of the Chinese art of face structure theory, an established acupoint location method, not standardizing the target face images may lead to lower rates of acupoint location in terms of precision and accuracy. Moreover, because of their special nature, most facial acupoint locations were unsuitable for the "bone standard" or "finger cun measurement" methods, so only the "anatomical landmark" method could be used. Therefore, in this paper, each correct acupoint location data point was obtained by acupuncture experts, and a variable precision method was proposed to improve weight computing. This method introduced weights for each expert, system error, and consistency. According to the method, weight of reliable data had been increased, so a higher reliability value could be got. Compared with other constant precision results, we could obtain a precise and accurate acupoint location value. Accurate and precise acupoint location for acupuncture treatments should be studied (a) "top down" as multi-component whole-system interventions, and (b) "bottom up" as mechanistic studies that focus on understanding how individual treatment components interact and translate into clinical and physiological outcomes.23 Such a strategy, incorporating considerations of efficacy, effectiveness and qualitative measurement, would strengthen the evidence base for such complex interventions as acupuncture. The alternative is to research efforts toward understanding the fundamental
(3) n
Y ' = ∑k i r i m iY i (4) i =1 n
δ x = ∑k i r i m i δ ix (5) '
i =1 n
δ y = ∑k i r i m i δ iy (6) '
i =1
X i , Y i is the actual measured value of the i-th expert acupuncture location acupoint, while δ ix , δ iy is the actual measured error of the i-th expert acupuncture location acupoint. The fusion "pointwise variable precision result" was calculated as: X' = ﹣ 39.5845, δ x' = 0.1865; Y' = 9.1599, δ y ' = 0.3954. This method was based on the original data inconsistency, and could not be achieved using the characteristics of the general weights calculation. Through correcting the system errors of various experts' acupoint location, reasonably quantifying the consistency of the locations, and obtaining fusion results of pointwise variable precision, measurements could more effectively be combined from different experts' acupoint location to increase the efficiency of measurement information. Patient welfare, clinical efficacy, and valid research results depend on accurate and precise acupoint location. A few years ago, the Australian scholar Mark Aird proposed that 22 improving the precision of acupoint location techniques could be seen as a three step process starting with measuring the required precision, then developing a more precise method, and finally identifying a procedure to measure whether an acupoint had been correctly located. Here, we have completed the first two steps.
DISCUSSION The accuracy and precision of acupoint location methods have important implications for research. Treatment outcome in clinical trials might be significantly altered if an inaccurate method of acupoint location was used. The task at hand was to instruct practitioners in the most precise and accurate method, rather than sacrificing experimental validity and clinical effectiveness for adherence to other methods. While the most likely consequences of imprecise acupoints are variability in treatment effects, there is also the potential for harm to patients and research participants if acupoints were not precisely located. Acupoint location descriptions and methods are vital considerations in research and practice. The field of acupuncture research has matured considerably, yet two provocative paradoxes have emerged.23 Acupuncture treatment components were made of non-specific JTCM | www. journaltcm. com
115
February 15, 2015 | Volume 35 | Issue 1 |
Yang XM et al. / Methodology
theories and practices of acupuncture, including the ability to precisely locate acupoints. Accurate and precise acupoint location has great significance for improving the curative effects of acupuncture. We collected information about acupuncture experts' acupoint locations on the face, and used an information fusion algorithm to obtain the most accurate and precise acupoint location values. Meanwhile, we developed a virtual application system to construct facial acupoint locations, to provide a platform for the learning of facial acupoint location, and to obtain the most accurate and precise acupoint location values on the face.
10
REFERENCES
15
1 2 3
4
5 6
7
8
9
11
12
13 14
Lao LX. Acupuncture situation and to explore clinical research. Zhen Ci Yan Jiu 2008; 33(1): 53-58. Liang FR, Ren YL, Tang Y. Acupuncture clinical evaluation. Zhen Ci Yan Jiu 2008; 33(1): 62-64. Yang XM, Xu LY, Zhong F, et al. Data mining-based detection of acupuncture treatment on juvenile myopia. Zhong Yi Za Zhi 2012; 32(3): 372-376. Liu WZ, Duan GJ. Acupuncture on the visual system and its mechanism of action research progress. Zhong Yi Yan Ke Za Zhi 1999; 5(3): 180. Chen XN. Chinese acupuncture and moxibustion. Beijing: Foreign Languages Publishing House, 2002: 60. Coyle MG, Mark A, Cobbin DM, Zaslawski C. The cun measurement system: an investigation into its suitability in current practice. Acupuncture Med 2000; 18(1): 10-14. Eckman P. In the footsteps of the yellow emperor: tracing the history of traditional acupuncture. San Francisco: Cypress Book Company, 1996: 66-70. Mark A, Deirder M, Carole R, Carole R. A study of the relative precision of acupoint location methods. J Altern Complement Med 2002; 5(8): 635-642. Aird M, Coyle ME, Cobbin DM, Zaslawski CJ. A study of the reliability of two methods of location acupuncture points. Acupunc Med 2000; 18: 15-21.
JTCM | www. journaltcm. com
16
17
18
19 20
21 22
23
116
Zhao Y, Zhang T, Lian QS. Infrared facial image Bell's palsy acupuncture location algorithm. Zhong Guo Ji Suan Ji Xi Tong Za Zhi 2012; 33(7): 1613-1619. Yan ZG. English-Chinese layer anatomical atlas of acupoints of the human body. Shanghai: University of Traditional Chinese Medicine Publishing House, 2005: 72. Wei LH, Wang W, Meng XM, et al. Principle in selecting acupoint and the prescription rule in acupuncture and moxibustion therapy. Tianjin Zhong Yi Yao Da Xue Xue Bao 2008; 27(2): 59-60. Zhao N. Face detection based on AdaBoost. Beijing: University of Beijing, 2005: 16-28. Xu LY. Source selection and dynamic feature extraction sequence level information fusion model and algorithm research. Shenyang: Northeastern University, 2001: 1-11. Zhon F, Liu N, Liu Y, Xu LY. New SST correction method from multi-satellite based on the coefficient of variation. Shanghai Da Xue Xue Bao 2011; 15(5): 463-466. Xu LY, Zhong F, Zhang PN. Conflict analysis of multi-source SST distribution. Shu Zi Feng Xi Yu Mo Xin Za Zhi 2012; 9(2): 338-342. Shi SX. Study on soft fusion strategy of multi-channel incertitude information in digital ocean. Shenyang: Northeastern University, 2005: 31-59. Wang J. Multi satellite monitoring SST data variable precision / reliability assessment studies. Shanghai: Shanghai University, 2011: 26-60. Wang XB. Study abroad meridian problems. Beijing: People's Health Publishing House, 1984: 9-15. Xiao YJ, Xiao P. Discussion acupoints on body surface area. Shandong Zhong Yi Yao Da Xue Xue Bao 1999; 23(1): 11-18. Zhu ZX. Engineering psychology. Shanghai: East China Normal University Publishing House, 1990: 10-15. Mark A. Variability in the precision of acupoint location methods. Sydney: University of Technology Sydney, 2005: 15-19. Langevin HM, Wayne PM, Macpherson H, et al. Paradoxes in acupuncture research: strategies for moving forward. Evid Based Complement Alternat Med 2011; 2011: 180805.
February 15, 2015 | Volume 35 | Issue 1 |