@2024 Afarand., IRAN

---

ISSN: 2251-8215    Sarem Journal of Reproductive Medicine 2017;1(3):105-112

Background
Data mining is a fast-growing interdisciplinary field that combines various domains, such as ‎databases, statistics, machine learning, and other related fields to extract valuable information and ‎knowledge that lies in the large amount of data [1]. One of the medical fields in which data mining ‎techniques can be used is infertility.‎
Previous Studies
‎... [2-4]. Of the commonly used treatments for infertility are assisted reproductive treatment (ART) that ‎have an effective role in treating infertility with various causes. Among the commonly used assisted ‎fertility techniques, ICSI, GIFT, IVF, IUI, and ZIFT can be mentioned [4].‎ Many studies have been conducted using data mining algorithms in the field of infertility, each ‎focusing on a specific factor. In a research, with the aim of evaluating the success rate of IUI treatment, ‎various factors such as the duration of infertility, couples' characteristics, sperm test status, etc. on the ‎data, concluded that the best treatment for infertility due to ovarian cancer is obtained by the IUI ‎method. [5].‎ In another study to determine the success rate of treatment with IVF, various factors affecting fertility ‎were examined. The results showed that ovarian response to ovulation stimulation and the number of ‎transferred embryos are important and effective factors in predicting IVF outcome [6].‎ In a study by examining the number of embryos transmitted in ART, it was found that by transferring 2 ‎or 3 fetus instead of 1 fetus, the probability of pregnancy increases significantly, on the other hand, ‎more increase of the numbers of the fetus is not significant and it increased the risk of multiple birth ‎‎[7].‎
Aim(s)
The aim of this study was to analyze the factors affecting the results of intrauterine insemination using ‎clustering.‎
Research type
This research is data mining. ‎
Research society, place and time

Sampling method and number

Used devices&materials
This study was conducted based on the well-known Cross-industry standard process for data mining ‎‎(CRISP, Figure 1). The processes followed in this study were:‎ Business Recognition: In this step of the analysis, the processes in the hospital were examined. This ‎stage is one of the most important stages of the research and during this project, it was repeatedly ‎returned to this stage to better understand the system of the hospital and to evaluate the results of data ‎mining with it.‎ Recognition of data: Patients refer to Sarem Hospital for various treatments including ICSI, IUI, GIFT, ‎ZIFT and IVF. Couples personal information and test results are recorded in their files. In this phase, ‎with consultation and assistance from the hospital team, all of the available data in the Sarem Hospital ‎were identified. This recognition includes all data records, data attributes, as well as familiarity with ‎the concepts of a variety of treatments (including types of tests).‎ Data preparation: one of the important phases of this study was the stage of data purification and ‎preparation for the data mining process, which was known as the preprocessing stage of the data. At ‎this stage, the data was selected for ease of research and purposefulness of the results, and the records ‎were collected from the hospital records, which differed in different fields of medicine and time, and ‎the data was collected and a survey on this category of data was started. In this phase, one type of ‎infertility technique was considered to limit the analysis. Data were collected from patients who ‎performed IUI surgery, and the clustering algorithm was performed on them and analyzed. Also, among ‎the fields in the database, some of the more effective parameters that were obtained by repeated ‎experiments in the field of previous research in this field and consulted by experienced people were ‎selected and used (Table 1).‎ After recognizing the hospital system and familiarizing with the processes that are carried out during ‎the treatment for patients, in the later stages, the recognition of the data and then the selection of the ‎required data from the massive amount of data to the stage of clearing and preprocessing the data was ‎done. At this point, the data was reviewed, categorized, deleted or modified. The result of data ‎purification was 400 cases between 1997 and 2009 from patients in Sarem Hospital.‎ In the next step, the selected clustering model was run on a prepared statistical sample to extract the ‎knowledge and patterns in this data.‎ Modeling: An uncontrollable (descriptive) algorithm was used to model data. One of the most ‎commonly used descriptive algorithm is the clustering algorithm used to analyze infertility data in this ‎study. Descriptive algorithms are used to discover similar patterns between different data sets. The ‎basic issue in the clustering algorithm used in the research is to distribute the data to a K group, so that ‎the data of each group be the same and the data of the different groups be inconsistent, which is called ‎‎"K-means" [9]. By using this non-explicit algorithm, a general analysis of data and a sort of data split ‎can be obtained. In using the clustering algorithm, a selection of an index is needed. In this research, ‎the Davis-Boulder Index was used. By reducing the value of this indicator, the number of clusters ‎similar to that indicator yields a better result.‎ To achieve a better analysis and better interpretation of the results, several clustering algorithms were ‎implemented with different indices on the data. This method, in addition to guidance for choosing the ‎best clustering index, helped to correct features and data. The clustering algorithms were performed ‎multiple times using different indices on the data, and it was finally determined that the Dunn and ‎Davis index were better indicators for implementing the algorithms on these data.‎ Dunn Index: This index is one of the most common indices in the assessment of the number of ‎clusters. Considering different ways of calculating the distance between two clusters and the size of the ‎cluster, different indices can be obtained from Dunn family. ‎ Davies-Bouldin Index: This index is a function of the ratio of the distribution of intra-cluster ‎dispersion to the separation of clusters [10].‎ Choosing the number of optimum clusters: The equations for the two above mentioned indicators ‎were implemented in the corresponding software, and the clustering of up to 7 clusters, were ‎performed on IUI data that took the following results: According to Dunn and Davies algorithms, if circles are drawn from the center of the clusters, it seems ‎that the clusters are intersecting (Chart. 1 and Fig. 2).‎ Therefore, the Dunn index did not provide a good result for this clustering, because, as mentioned, this ‎index is based on the distance between the cluster and the diameter of the cluster, that the higher the ‎number, the better it is. In comparison, The Davies index focuses on the density-to-distance ratio, and ‎does not stand on distance alone, resulting in less sensitivity to the cross between clusters. The ‎clusters seem to intersect each other while it is not true (Figure 2).‎ Considering the optimal number of each of these indices, the Dunn index considered the 2 clusters as ‎the best and Davis index considered 6 clusters as the most optimal number of clusters, and ‎considering the discussion, 6 clusters were considered as the optimum number of clusters. Since it is ‎possible to be exposed to local optimal in K-means algorithm, we used another algorithm called EM ‎‎(Expectation Maximization) [11] to insure the result of the K-means algorithm and the DB index. The ‎results of the EM algorithm also provided us with 6 clusters. In this algorithm, for each sample, a ‎probability distribution is assigned that determines the probability of each cluster. This algorithm can ‎make a decision about the number of clusters appropriately implemented using the internal cross ‎validation [11] algorithm that it implements. The algorithm was implemented in the WEKA program. ‎As stated, the cross validation algorithm detects the number of appropriate clusters, which is done as ‎follows.‎ The two algorithms presented the same result with two different approaches, but in the final stage of ‎certainty assurance of the number of clusters, the results of these two algorithms were checked with ‎experts in this area at Sarem Hospital in order to ensure the accuracy of the implementation of these ‎algorithms (Fig. 3).‎
Findings by Text
After 7 times implementation of the K-means algorithm that was selected for data clustering, and ‎considering the Davis-Bouldin index selection for choosing the number of optimum clusters, 6 clusters ‎were selected and the dispersion of patients in each cluster was determined (Table 2).‎In following, each of the six selected clusters is analyzed. The results of cluster 1 are shown in Table 3. ‎Six samples of samples in the top cluster were not recorded in the database for their success or failure ‎‎(Table 4).‎The results obtained from the second cluster are shown in Table 5 and the results of the success and ‎failure in this cluster are also described in Table 6.‎The results obtained from the third cluster are also described in Table 7, and the results of success and ‎failure are presented in Table 8.‎Based on Table 8, 1 sample of samples in the top cluster was not recorded in the database for their ‎success or failure. The results obtained from the fourth cluster in Table 9 and the results of the success ‎and failure of this cluster are also listed in Table 10.‎The results obtained from the fifth cluster are also described in Table 11; the results of the success and ‎failure in this cluster are also described in Table 12.‎The results of the sixth cluster are also described in Table 13. According to Table 13, 5 samples of ‎samples in the top cluster were not recorded in the database for their success or failure. The results of ‎the success and failure in this cluster are also in Table 14. Based on Table 14, 6 samples of the samples ‎in the top cluster were not recorded in the database for their success or failure.‎
Main comparison to the similar studies
In a similar study to the present study, but on IVF, the success rates of treatment with IVF and various ‎factors affecting fertility were evaluated. The findings indicated that ovarian response to ovulation ‎stimulation and the number of transferred fetus are important and effective factors in predicting IVF ‎outcomes [5]. In 2008, another study was conducted on the probability of gestation with the number of ‎embryos transmitted. By examining the number of embryos transmitted in ART, it was determined by ‎transferring 2 or 3 fetuses instead of 1 fetus, the probability of pregnancy increases significantly, and ‎by increasing more fetus, the increase in the rate of the probability is not significant and does not cost ‎risk of multiple birth. In that study, using Bayesian‏ ‏network techniques, the probability of multi birth ‎was investigated [7].‎ In other areas, such as prediction of IVF treatment, research has been done, such as a research in 2010 ‎that found 39 features based on sampling, characterization and data cleansing, to enhance accuracy, the ‎combination of decision tree and genetic algorithm was used [12]. Other systems that can be ‎mentioned in the field of infertility are the system for predicting the appropriate type of IVF treatment ‎using CBR, called the TA3IV system. The goal of the study is to provide CBR with stored experiences a ‎possible way to increase the success rate for IVF. After knowledge is a system based on a good ‎population of samples, the feasibility of exploiting the knowledge used out of a large amount of data is ‎found. In this study, the TA3IV system, which is based on a Danish-based CBR system, has been ‎described. This system can be a sponsor with doctors [13]. In a study, 2450 couples were studied, ‎which was also a good basis for selecting features in that study, but only a statistical survey of patients ‎was noted. Among the selected features were BMI, type of infertility, infertility length, age, etc. [14]. ‎The main difference in this research is that despite the field of success and failure in performing the IUI ‎operation, they first were removed from the feature, and only the status of the patients was evaluated, ‎because the main consideration was obtaining the general knowledge of the patients in the hospital and ‎then following the analysis or the data, the success or failure of each patient could be evaluated based ‎on the their clustering , and this can be used as base in the future to create a decision-making system.‎
Suggestions

Limitations

Conclusions
Factors such as age, body mass, type of infertility, cause of infertility, etc. can determine the success ‎rate of the IUI method.‎ ‎
Acknowledgments
Dr. Saremi, the dear chairman of the hospital, and Dr. Salehian, as well as the competent and highly ‎qualified staff of the Sarem Medical Women's Specialized Hospital, who helped us in the various stages ‎of this study, especially the work of Ms. Ghafari and Ms. Shami are thanked and appreciated. ‎
Conflict of interest

Ethical Permissions

Funding Sources

TABLES and CHARTS

Show attach file