Data Mining-Based Detection of Acupuncture Treatment on Juvenile Myopia

CLINICAL OBERVATION Data mining-based detection of acupuncture treatment on juvenile myopia

Xuming Yang, Lingyu Xu, Fei Zhong, Ying Zhu aa Xuming Yang, College of Acupuncture and Massage, Shang- consistent with the understandings of the ophthal- hai University of TCM, Shanghai 201203, China mic circles. The duration of using the Internet and Lingyu Xu, Fei Zhong, College of Computer Engineering watching TV every day was the main factor that af- and Science, Shanghai University, Shanghai 200072, China fected vision. Acupuncture feelings and therapeu- Ying Zhu, Department of Ophthalmology, Shanghai of TCM tic effect have a strong correlativity. A good or hospital, Shanghai 200071, China Supported by National Natural Science Foundation grant above experience's score of acupuncture could NO.40976108 and Public Projects of Science and Technolo- slow the progression of juvenile myopia. gy Ministry grant NO. 201105033 Correspondence to: Prof. Lingyu Xu, College of Computer CONCLUSION: Collecting data from patients with Engineering and Science, Shanghai University, Shanghai juvenile myopia by using data mining can extract 200072, China. [email protected] hidden potential rules and knowledge from the re- Telephone: +86-21-56337465, +86-13122852765 search evidence. The decision support can be pro- Accepted: June 25, 2012 vided to improve the doctor's clinical acupuncture treatment effects.

© 2012 JTCM. All rights reserved. Abstract OBJECTIVE: We applied data mining techniques to Key words: Acupuncture therapy; Myopia; Algo- the study of acupuncture as a treatment for juve- rithms; Data mining nile myopia, with the aim of identifying hidden patterns in the data. INTRODUCTION METHODS: Fifty patients with juvenile myopia According to recent surveys at home and abroad, Chi- were selected and treated with acupuncture, and na has one of the highest incidences of myopia, and data mining was used to analyze the effects of the incidence of juvenile myopia is increasing.1 Previ- treatment and the influence of behavioral vari- ous studies of myopia have been limited to modern evi- ables. Clustering analysis was used to divide myo- dence-based ophthalmology. So far, modern medicine has not only failed to find a cure for myopia, but it has pia patients into two classifications before acupunc- also failed to find a feasible way of controlling its devel- ture treatment. Artificial neural network BP algo- opment.2-4 Therefore, it is important to explore the ef- rithm was adopted to analyze the roles of different fectiveness of treatments and prevention methods for factors in changes in diopters. An association algo- true myopia, especially high myopia, which has impor- rithm was used to analyze factors associated with tant medical and social consequences.5-8 Acupuncture the subjective experience of acupuncture and aver- can treat a variety of eye disease; the theoretical basis of age diopter. this treatment is the relationship of the eyes to the 12 meridians.9-10 In acupuncture, clinical evidence-based RESULTS: The two classification results were fully diagnosis and treatment decision-making analysis have

JTCM | www. journaltcm. com 372 September 15, 2012 | volume 32 | Issue 3 | Yang XM et al. Data mining-based detection of acupuncture treatment on juvenile myopia focused on two aspects. The first is based on the pa- Data mining algorithms tient's signs and symptoms for diagnosis, resulting in a Knowledge of the association is to reflect the knowl- dialectic treatment decision. The second relies on analy- edge of the dependence or association between events sis of the known rules of acupoints and efficacy of acu- in a data set. The important feature of association rules puncture, thorn moxibustion, treatment time, and oth- is an "association natural combination", which found er relevant factors, to decide on the optimal treatment that a subset that existed mode of all the attributes is solution.11 The foreign ophthalmic industry has used very useful.16,17 The purpose of clustering is reasonable decision tree analysis to find patterns regarding risk fac- according to certain rules of classification or clustering, tors for eye deterioration in large datasets.12 The domes- and with the explicit or implicit method to describe tic ophthalmic industry using Chinese medicine re- the different categories. By digging the hidden implicit gards ophthalmology as "one of the hardest parts in knowledge in the treatment of information, it can pro- TCM", and believes in the use of data mining methods vide decision support for the improvement of the doc- to study the effectiveness of TCM in ophthalmology.13 tor's clinical acupuncture treatment. An artificial neu- In this study, we used the association rule learning ral network BP algorithm can approximate any nonlin- method of data mining using data acquired during acu- ear curve. To identify which factors are associated with puncture treatment of juvenile myopia. We aimed to the patient's average diopter change, the network is giv- find hidden associations within the data, in terms of en a set of inputs and corresponding outputs and deter- treatment options, treatment efficacy, and influence of mines the parameters of each neuron by analyzing the co-existing medical conditions. We verified through ex- relationship between the input and output.18 In previ- periments the effectiveness of acupuncture treatment ous research, multiple stepwise regression analysis has in juvenile myopia, and explored new information on been applied to the quantitative analysis of factors in- how to further optimize treatment and prevention mea- fluencing the response to treatment in patients receiv- sures. ing acupuncture for juvenile myopia. Each patient has a unique optimal equation, and by studying the model, we can obtain the weights of factors determining visual METHODS AND ALGORITHMS acuity.19-22 Acupuncture doctors need as much information as possible about the patients and treatment, so that they can adjust the acupuncture treatment or Patient selection and data collection adopt other effective measures based on the relevant Fifty patients with true juvenile myopia were selected factors in the optimal equation of each patient. This according to the textbook Chinese medicine Ophthal- provides a more targeted way of selecting an appropri- mology 14. The treatments were chosen based on the fol- ate acupuncture treatment program early on in the clin- lowing principle: promoting flow of Qi and blood cir- ical course, and also provides a more reliable means of culation, relaxing muscle and tendons, and improving tracking the progress of treatment and assessing the the eyesight. Stimulation of eye acupoints can regulate therapeutic effect. the functions of meridians of eyes and zang-fu organs.15 The acupuncture points Cuanzhu (BL 2), Taiyang (EX-HN5), Qiuhou (EX-HN7) and Sibai (ST 2) are DATA MINING ANALYSIS located around the eye area, and their stimulation can clear blockages and regulate the blood flow and meridians. Hegu (LI 4) is the original acupoint of the large in- Questionnaire on the impact of juvenile myopia testine meridian of hand-yangming. Needling can agi- during acupuncture treatment tate Qi and blood in the face near the eye area, promot- During acupuncture treatment of juvenile myopia, pa- ing Qi and blood flow and improving eyesight. Baihui tients were required to complete "the focus on object (GV 20) belongs to the Dumai (Governor Vessel) and tracking questionnaire". Questions were related to meets with the bladder meridian of foot-taiyang. Thus, whether the parents suffered from myopia, the daily acupuncture stimulation of Cuanzhu, Taiyang, Qi- burden on the eye, amount of time spent on the Inter- uhou, Sibai, Baihui and Hegu has an effect in the treat- net and watching television each day, reading time, ment of juvenile myopia. We collected clinical data at writing time, reading posture of the patients, usual 1, 2, 3 and 4 months from the start of the treatment, time spent on homework, the subjective experience of using several eye examination tests, including visual acupuncture, whether the patient likes sweets and fried acuity, intraocular pressure, refraction, horizontal cor- food, and other questions. The extracted information neal curvature, vertical corneal curvature, axial length, was encoded in the table as follows. Parental myopia: " lens thickness and depth of anterior chamber and vitre- 0" indicates "no myopia", "1" indicates "myopia"; ous matter. Other indices based on self-designed "acu- learning environment: "0" indicates "good", "1" means puncture treatment of juvenile myopia focus on the ob- "poor "; reading position: "0" indicates "good", "1" ject of study follow-up survey table" were collected as means "poor"; eating habits: "0" indicates "good", "1" the source data for data mining. means "poor"; time spent on the Internet and watch-

JTCM | www. journaltcm. com 373 September 15, 2012 | volume 32 | Issue 3 | Yang XM et al. Data mining-based detection of acupuncture treatment on juvenile myopia ing TV: the number of hours/day, such as "1", "2", patients were divided into two classifications, with 24 etc.; subjective experience of acupuncture: rated on a patients in classification 1, and 26 patients in classifica- scale from 0 to 4, where "4" means "very good", "3" tion 2. means "good", "2" means "effective", "1" means "neu- In the Microsoft SQL Server 2008 Analysis Services tral" and "0" means "poor"; time spent reading/writ- (SSAS), before acupuncture treatment, cluster analysis ing: the number of hours/days, such as "1", "2", etc. algorithm can get classification's characteristics as fol- Assessment of the learning environment and reading low. The main variables impacted by myopia were: posture was based on the Shanghai Municipal Educa- "reading position" (94% ), "parents' myopia" (80% ) tion Commission promulgated "two requirements for and "eating habits" (72%). The classification and com- do" and "another two requirements for donot" of "pro- parison is shown below. At the same time, the classifica- tection eyesight, the prevention of myopia and hygiene tion and comparison shows a daily "eye time" of two knowledge", to determine "good" and "bad". hours, Internet/TV time of two hours and parents with myopia favored classification 1, while a daily "eye Cluster analysis of factors associated with juvenile time" of three hours, Internet/TV time of three hours myopia before acupuncture treatment and no myopia in parents favored classification 2. Fur- The patients were classified according to an ophthal- ther analysis showed that patients with juvenile myopia mologist recommendation obtained before the acu- could be divided into two groups: one without the ge- puncture treatment.23,24 We were helped by Chinese netic component and including patients with long-du- ophthalmologists to understand the patients' condition ration (>3 h) eye time, and one consisting of patients before the acupuncture treatment, taking acupuncture with eye time of ≤2 h but with myopic parents. These treatment for targeted prescription, and doing targeted results are in agreement with the current views of the therapy. Maximum expected value of the soft cluster- ophthalmic industry.25 ing method, a closed curve as the clustering criteria set in each dimension, and calculate the mean and stan- Effect of behavioral factors on the average diopter dard deviation. If a point falls on the closed curve, change during acupuncture treatment then there is a certain probability of the point belong- In the sample of 50 juvenile myopia patients, four ing to the given classification. In addition, because the months of acupuncture treatment was associated with classification of the corresponding curve is not unique, an average diopter change in some patients but not oth- some points are likely to fall into more than one classi- ers. However, during acupuncture treatment, the pa- fication curves; these were assigned a probability value. tients had their own lives, learning habits and behav- We used "ID" as the key and "parents myopia", "reading posture", "eating habits", "eye time" and "Internet iors, and all these factors may affect the average diopter TV" as the input. The "average diopter" was the pre- change. The data concerning "Internet/TV", "parents' dictable variable. Because the "learningΔ environment" myopia", "eating habits", "reading position" and "eye was classified as "good" in all cases, this variable was time" are shown in Table 1 together with the left and not included. The clustering result showed that the 50 right eye average diopter change. Table 1 Living habits and average Δ diopter Read- The av- Date L R Parents Eating Eye Internet ID NO Gender LID ing po- erage of birth Diopter△ Diopter△ myopia habits time TV sition diopter△

1 NO.1 M 1998.11.26 1 0 -0.25 1 1 1 3 0 -0.125

2 NO.2 F 1999.12.1 2 0 -0.25 1 0 1 3 1 -0.375

3 NO.3 M 1997.11.20 3 0 0 0 1 1 4 0 0

4 NO.4 F 1988.2.7 4 -1.25 -0.75 0 1 1 3 1 -1.00 ... … … … … … … … … … … … … Using a BP algorithm of artificial neural networks,26 is, the input attribute "Internet/TV=0" favored output with ID as the key, "Internet/TV", "eye time", "par- attribute "average diopter=–0.375" with a score of ents' myopia", "eating habits" and "reading position" 100% . The inputΔ attribute "eye time=3" favored out- as the input, and "average diopter" as a predictable put attribute "average diopter=–0.375" with a score output. Using the output attributeΔ "average diopter" of 67.9% . Similarly, theΔ input attribute "eye time=4" and Value 1=– 0.375 and Value 2=0 as anΔ example, favored output attribute "average diopter=0" with a since there are different input attribute values associat- score of 63.52%, and "Internet/TV=3"Δ favored "aver- ed with the two output attribute values, we select the age diopter=0" with a score of 39.34%. As a result, first two input attributes that most strongly affect the we minedΔ the data generated by analysis of SSAS as output attribute as the main influencing factors. That shown below Table 2.

JTCM | www. journaltcm. com 374 September 15, 2012 | volume 32 | Issue 3 | Yang XM et al. Data mining-based detection of acupuncture treatment on juvenile myopia

Table 2 Relative importance of the contribution of input variables to average diopter change Score Output attribute value (The average diopter) △ Input attribute value -1.75 -1.25 -1 -0.875 -0.75 -0.375 -0.25 -0.125 0 InternetTV =3 63.42 39.34 InternetTV =2 100 41.1 54.12 InternetTV =1 35.31 InternetTV =0 100 Eye time =4 100 63.52 Eye time =3 51.77 29.23 67.9 Eye time =2 45.13 Eye time =1 Eating habits =0 49 97.75 Eating habits =1 56.44 Parents myopia =0 73.4 Parents myopia =1 Reading position =0 100 Reading position =1

Table 2 shows how when a value is specified for the in- ticular when the experience of acupuncture is "good" put variables, it affects the score of the output vari- or above. Association analysis thus showed that there ables. For the input attribute value "Internet/TV", was a strong association between the experience of acu- there were seven occurrences of scores, for "eye time" puncture and therapeutic effects. The subjective experi- there were six scores, for "eating habits" three scores, ence of acupuncture can be used to determine whether and for "parents myopia" and "reading posture" there a patient needs to receive individualized treatment, was one score. Therefore, the main factor affecting the such as scalp or ear acupuncture, for a more targeted ef- diopter change in juvenile myopia is time spent on the fect. Internet and watching TV, and this is followed by "eye time". Furthermore, visual acuity was also affected by a preference for fried food and sweets, the presence of parental myopia, and a bad reading posture. At this SUMMARY point, it is recommended that patients with juvenile myopia combine labor with rest and watch their diet, We studied 50 cases of juvenile myopia during four to protect their own vision. months of the acupuncture treatment, by using myopia detection data, data mining and knowledge acquisi- tion. Before the acupuncture treatment, cluster analysis Association analysis of the subjective experience of resulted in the classification of myopia patients into acupuncture in visual acuity change two groups, which is fully consistent with the views of After the end of the four courses of acupuncture treat- the ophthalmic industry. After acupuncture treatment, ment, the doctors hope to be able to know the role of we used an artificial neural network BP algorithm to acupuncture in the treatment of myopia. We used an analyze the diopter change during treatment and to an- association algorithm with the "ID" as the key, "experi- alyze the influence of different behavioral factors, find- ence of acupuncture" as input and "average diopter" ing that the duration of using the Internet and watch- as a predictable variable. When the minimumΔ probabil- ing TV every day was a main factor that affected vi- ity was set to 0.64 and the lowest importance to 0.26, sion. Therefore, patients should be persuaded to avoid it resulted in three association rules. According to de- television and Internet as far as possible to protect their pendency network (from strongest to weakest link): ex- vision. We also found that there was a strong correla- perience of acupuncture<0.9196 à average diop- tion between the subjective experience of acupuncture ter=-0.6918 to -0.0592; experience of acupuncture=Δ and the effect of treatment, whereby a good experience 2.7499-3.1599 à average diopter≥-0.0592; experi- of acupuncture could slow the progression of juvenile ence of acupuncture≥3.1599Δ à average diop- myopia. To enable patients to have a good experience ter≥-0.0592. This shows that when the experienceΔ of of the acupuncture treatment, the optimal equations acupuncture was neutral (score~1), the diopter in- gained by a multiple stepwise regression analysis algo- creased relatively quickly, whereas when the experience rithm can be used to find the main impact indicators.19 of acupuncture was good (score~3), the diopter in- This enables the selection of the best acupuncture pre- creased more slowly. This indicates that acupuncture scription or individualized treatment. However, if pa- treatment prevents the development of myopia, in par- tients are genuinely unable to reach the "good" level,

JTCM | www. journaltcm. com 375 September 15, 2012 | volume 32 | Issue 3 | Yang XM et al. Data mining-based detection of acupuncture treatment on juvenile myopia they should be persuaded to terminate the treatment, thalmology 2008; 18(6): 366-368. so as not to waste human and material resources. 14 Zeng QH. Chinese medicine Ophthalmology. Traditional Chinese Medicine Press, 2009. 15 Diao CY, Huang JD. Huang acupuncture holistic therapy Prevention of myopia experience in children. Chinese REFERENCES Medicine 2007; 25(2): 227-228. 1 Chen YQ. Chinese medicine combination therapy for 16 Sun YH, Sun YJ. Association rules mining Apriori Algo- treatment of juvenile myopia observe the effect. Sichuan rithm. Academic journals of Jilin Normal University 2009; Traditional Chinese Medicine 2009; 127(12): 110-111. 11(4): 82-84. 2 Saw SM, Shih Yen EC, Koh A, Donald T. Interventions 17 Xue YS, Li YQ, Shi Rui, Wang Ke. Juvenile myopia diop- to retard Myopia progression in children: an evi- ters and axial length and intraocular pressure correlation dence-based update. Ophthalmology 2002; 109(3): analysis. International Journal of Ophthalmology 2008; 8 415-426. (9):1847-1849. 3 Li WS. To carry out evidence-based eye of scientific re- 18 Chan CP. Data warehouse and data mining.Beijing: Tsing- search in China imperative. Journal of Ophthalmology hua University Press 2009: 188-193. 2004; 40(2): 75-77. 19 Yang XM. Acupuncture treatment of juvenile myopia ef- 4 Saw SM, Wong TY. Is it possible to slow the progression fect model of mining technology. Shanghai University, of myopia. Ann AcadMed Singapore 2004; 33(1): 4-6. Master Thesis 2011. 5 Qu Jia. Adhere to the prevention and treatment of myo- 20 Yang LN. Study and Application on the Occurrence Pre- pia research the right direction. Journal of Ophthalmology diction Model of paddy Stem Borer (Scirpophaga Incertu- 2003; 39(6): 321-322. las) and Liriomya Huidobresis (Blanchard).University of 6 Zheng RL, Zhai DL. Chinese medicine can contribute to the prevention and control of juvenile myopia. Chinese Electronic Science and Technology of China ,Doctor The- ophthalmology 2006; 13(12): 8. sis 2009. 7 Wu YL. Young people how to protect their eyesight. Chi- 21 David R, Zangwill LM, Tessler Z, Yassur Y. The correla- nese ophthalmology 2007; 13(9): 1260-1261. tion between intraocular pressure and refractive status. 8 Zhang SK, Wang SK. The Present Situation of Chinese Arch Ophthalmol 2001; 103(3): 1812-1815. medicine treatment of juvenile myopia. Chinese folk medi- 22 Quinn GE, Berlin JA, Young TL, Zivlan S, Stone RA. As- cine 2006; 14(3): 63-65. sociation of intraocular pressure and myopia in children. 9 Liu WZ, Dian JG. Research progress of the effect and Ophthalmology 1995, 102(6): 180-185. mechanism of acupuncture on the visual system. Chinese 23 CAI YC, Chen LC. Clustering Algorithm Research. Sci- ophthalmology 1999; 3(7): 180. 10 Shang XJ. The acupuncture End treatment of myopia of ence and Technology Information Development and Econ- the bone hole 123 Cases. Hebei Traditional Chinese Medi- omy 2007; 17(1): 145-146. cine 2008; 30(12): 1310-1312. 24 Shen R. Exploration and research of data mining tech- 11 Liang FR, Ren Yl, Tang Yong. Acupuncture clinical effica- niques clustering algorithm. Shanxi Science and Technolo- cy evaluation studies. Acupuncture Research 2008; 33(1): gy 2009; 3(3): 90-91. 62-64. 25 Li YY. The causes and treatment of juvenile myopia. Hu- 12 Susan P, Imberman, Irene Ludwig, Sarah Zelikovitz. Us- bei College of Traditional Chinese Medicine 2008; 10(2): ing Decision Trees to Find Patterns in an Ophthalmology 50-51. Dataset, Proceedings of the Twenty-Fourth International Florida Artificial Intelligence Research Society Conference 26 Ding XL, Yang Tao, Zhou JH. The use of artificial neural 2011. network analysis of the impact of the disease factors–for 13 Yao XL, Peng QH. Data mining technology envisaged in example Hypertension. Medical Informatics 2009; 22(1): the application of Chinese ophthalmology. Chinese oph- 4-7.

JTCM | www. journaltcm. com 376 September 15, 2012 | volume 32 | Issue 3 |