Journal of Engineering, Computers & Applied Sciences (JEC&AS) Volume 2, No.9, September 2013

ISSN No: 2319-5606

Applications of Artificial Neural Network for IVF Data Analysis and Prediction Dr. M. Durairaj, Assistant Professor, Department of Computer Science, Engineering and Technology, Bharathidasan University, Tiruchirappalli, Tamilnadu, India P. Thamilselvan, Research Scholar, Department of Computer Science Engineering and Technology, Bharathidasan University, Tiruchirappalli, TN, India

Abstract This paper aims to predict the success rate of in-vitro fertilization (IVF) using Artificial Neural Network (ANN). Artificial Neural Networks are founds very useful for a number of medical diagnosis applications [1]. In this work, the ANN is used for processing the patients IVF data and assessing the possible success rate of the treatment which could help the gynecologist to suggest the infertility patients who undergo fertility treatment for baby. In recent years, the number of infertile couples seeking infertility treatment rapidly increases due to the increasing awareness of test tube baby treatments. The increased success rate of infertility treatment through IVF and ICSI (intra-cytoplasmic sperm injection) makes the people to consider this option to have babies. The treatment of having babies though IVF / ICSI is a costly affair and there is no reliable methodology to assess the success rate of the treatment, and success rate differs from patients to patients. There are a number of factors affecting the success of the particular treatment, including male and female factors and various IVF test results. Even the psychological factors of the couples play major role in deciding the success of the treatment, and multiple cycles of the treatments increase the cost as well as affecting the patients’ health and increases the stress level. In this work, the efforts of applying artificial neural network for predicting the success of the IVF treatment for individual couples who undergo fertility treatment is carried out. The data used in this work contain the information of various tests and medical examination results of the couples such as endometriosis, tubal factors and follicles in the ovaries, and the physiological factors such as stress level factors. Keywords Artificial Neural Network (ANN), In Vitro Fertilization (IVF), Test tube baby, Fertility Rate and Back Propagation Algorithm.

1. Introduction In recent years, the application of Artificial Neural Network found in a number of fields including medical, biological, engineering, robotics, social and economic applications [2]. In medical applications, the Artificial Neural Networks are used to increase the accuracy of medical diagnosis for assessing the disease of the patients. The researches on the application of artificial neural network in diverse fields have been increased success in recent years [3]. The usage of Artificial Neural Networks is found in diagnostic systems, image processing, signal processing, clinical medicine and biomedical applications. The Artificial Neural Networks have been successfully applied in predicting the fertility rate of livestock, since it has a great economic importance for breeding cattle [4]. The Artificial Neural Network can effectively replace traditional statistical prediction methods with more accuracy [5].

www.borjournals.com

In this work, the patients‟ data include female age, duration of infertility, BMI (body mass index), previous pregnancy, previous surgery, endometriosis, tubal causes, ovulatory factor, sperm concentration, sperm vitality, number of oocytes retrieved, number of embryos transferred, previous history of miscarriages, and psychological factors. Artificial Neural Network can be a useful technique for the prediction and classification of medical data [6] [7]. This paper proposes a methodology of using Artificial Neural Network for data analysis and assessing the success rate of individual infertility couples about to undergo IVF / ICSI treatment for having a test tube baby [8] [9].

2. Applications Of Artificial Neural Network 2.1 Data Set The medical data sets used in this work have been collected from IVF research center, Fertility clinic and Maternity Hospitals. The collected data consists of 250 patients‟ records with 27 attributes

Blue Ocean Research Journals

11

Journal of Engineering, Computers & Applied Sciences (JEC&AS) Volume 2, No.9, September 2013

of lab results subject to the preprocessing as suitable for input to ANN prediction. The data set from the infertile couples who successfully conceived after IVF treatment have been recorded for training the network, that can be used for prediction.

2.2 Construction of Artificial Neural Network

ISSN No: 2319-5606

problems. The ANN can be constructed as a computational model to describe relations between inputs and outputs using mathematical relevance between the attributes of data set. In this work, ANN model construction is carried out using a software tool Neuro Solution, in which a multilayer perceptron with a single hidden layer, input and output layer as shown in Fig.1.

2.3 Prediction Model Development The block diagram of the overall prediction model has been illustrated in Fig. 2, where input is various IVF test parameters and output is the success rate. The selected input data are properly pre-processed and normalized as a suitable input pattern to Artificial Neural Network. After the development of artificial neural network model, the data prepared and composed into proper format data as training, data as test and data as cross validation and data as a target.

BMI- Body Mass Index, Endo- Endometrices, DI- Duration of Infertility, TI- Tubal Infertility, SC- Sperm Concentration, ORFig 1: A Simple Model of ANN No. of OOCYTES Received, ET-EmbryosTransferred, IVF-T– InVitroFertilization Treatment

An Artificial Neural Network (ANN) is a computational model based on the structure and functions of biological neural networks. The information that follows through the network affects the structure of the ANN because a neural network changes based on the input values. Artificial Neural Networks are considered the nonlinear statistical data modeling tools where the complex relationships between outputs and inputs are modeled and patterns are found. The types of ANN based on their structure are Single layer perceptron, multilayer perceptron, Kohonen feature map, Hopfield net, Back propagation algorithm and multilayer perceptron (MLP), in which MLP is recognized as the best ANN structure widely used in classification and prediction based on learning from the examples [10]. Artificial Neural Network is an interconnected group of artificial neurons that used to construct a computational model for information processing and derive solutions from complex computational

www.borjournals.com

Fig 2: Artificial Neural Network (ANN) based Prediction system

2.4 Network Training Classification The Artificial Neural Network is used for the prediction of fertility success rate based on the IVF data [11]. A different process is involved in the Artificial Neural Network are: (1) selecting data for training, (2) selecting data for cross validation, (3) selecting data for testing, (4) analyzing and transforming the selected data, (5) network construction and training. A properly trained neural network is capable of generating the information on the based on IVF data. During the training phase if the sample data contain noisy information. To train an artificial neural network, a suitable training, cross validation and test data are selected. The neural network is trained with the training data, and check with test data. The ANN will find the desired output-actual output map from

Blue Ocean Research Journals

12

Journal of Engineering, Computers & Applied Sciences (JEC&AS) Volume 2, No.9, September 2013

the training set. In this work, the neural network would have three layers, such as the input layer, output layer and hidden layers. The Artificial Neural Network layers, whose structure is shown in Fig: 1 and it has the following characters: a) Input layer: 8 Nodes as selected parameters for training and testing; b) Hidden layer: The number of nodes is used in the hidden layer may be varied as per the validation data; c) Output layer: output layer of the constructing neural network is for producing success rate of IVF treatment.

3. RESULT AND DISCUSSION 3.1 Learning Coefficient Optimization for Artificial Neural Network

ISSN No: 2319-5606

configurations (ANN), based on their predictive performance. It was measured by calculating the MSE (Mean Square Error), NMSE (Normalized Mean Square Error) and MAE (Mean Absolute Error) are defined as: The formula for the Mean Squared Error (MSE) is:

….. (1) Where P = number of output processing elements N = number of exemplar in the data set yij = network output for exemplar i at processing element j dij= desired output for the exemplar i at processing element j The formula for Normalized Mean Squared Error (NMSE) is:

…… (2) Where P = number of output processing elements N = number of exemplars in the data set MSE = mean squared error dij = desired output for exemplar i at processing element j Best Networks Epoch # Minimum MSE Final MSE

Training

Cross Validation

999

19

1.34012E-12

0.351037601

9.65898E-12

0.411017223

Fig 3: Mean Square Error vs. Epoch for Training and Cross Validation The artificial neural network is used to evaluate the training and cross validation. The training is carried out using training rule based on the Back propagation algorithm (error back propagation) [12]. Back Propagation algorithm is a fast algorithm comparing to other algorithms. The Back Propagation algorithm is based on the gradient based method. It is specially used for minimizing the MSE (Mean Square Error) of a neural network. The prediction errors (MSE) between the predicted data and actual data are analyzed. (As shown in the Fig: 3). The optimal artificial neural network configuration is selected from various neural network

www.borjournals.com

The formula for the Mean Absolute Error (MAE) is: ….. (3) The training is using a training rule, based on the Error back propagation algorithm that is a more effective method for training. During the training, the weights and network are adjusted to minimize the error network function. Training the network multiple times produces the lowest error. The plot shows the Mean Squared Error (MSE) of the network for each epoch of data. The epoch number is shown on the X-axis and the Mean Squared Error, on the Y-axis. The Mean Squared Error (MSE) of the training set is shown in red color. The Mean Squared Error of the Cross Validation set is shown in blue color (Fig. 3). Training Network is constantly decreasing slope of training Mean Squared Error. The training set learning curve is decreasing means, and then the network is still running. If the training set learning curve is increasing (up and down) means, then the network will not train well. The cross validation learning

Blue Ocean Research Journals

13

Journal of Engineering, Computers & Applied Sciences (JEC&AS) Volume 2, No.9, September 2013

curve starts to increase, and then the network training data and the network should be stopped. In this work, the trained network contains the 999 epochs approximately (see Figure 3).

3.2Sample Predicted Actual Network Output Once the network has trained, the network is ready to run tests. After completion of proper training, validation and testing, the Artificial Neural Network predicts the actual output (as shown in table 1). The desired output is the value obtained from recorded field data, which is the actual result of treatments carried out on infertility patients. The actual network output is the output of neural network. Patients No

Desired Output

Actual Network Output

001 002 003 004 005 006 007 008 009 010 011 012 013 014 015 016 017 018 019 020 021 022 023 024 025 026 027 028 029 030 031 032 033 034

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 1 1 1 0 1 1 1 1 0 1 0 1 0 1

-1.912E-05 -2.272E-05 -9.909E-07 -1.248E-05 -5.224E-05 6.8822E-05 8.0355E-05 -1.245E-05 -1.198E-05 -1.245E-05 -1.466E-05 -4.32E-06 -1.348E-05 -3.658E-05 1.5117E-06 -9.302E-05 0.99988895 0.50074115 0.00114295 0.55270079 1.015785 1.05297999 1.04702774 2.909E-06 0.9999346 1.00000666 0.9214552 0.86304731 -1.568E-05 1.00523945 0.01357123 0.99994848 0.17453354 0.99898372

Table 1: Sample Predicted Actual Network Output

ISSN No: 2319-5606

3.3Comparison of Desired Output and Actual Output The comparison of desired output and actual network output is shown in below (Fig 4). The desired output is shown in line and the actual output is shown in dash line (Fig 4). In This Figure actual output line is travelling the near of desired outputs. So finally this paper shows the most perfect output

Fig 4: Comparison of Desired Output vs. Actual Network Output

3.4 Result This Table (Table 2) describes the Mean Square Error (MSE), Normalized Mean Square Error (NMSE), Mean Absolute Error (MAE), and correlation coefficient (r) for predicting the success rate of IVF data. The size of MSE, NMSE, and MAE can be used to describe how well the actual output fits to the desired output. When the correlation coefficient „r‟ is 1 means the network is predicting the perfect output. Since r is 0.49, the correlation between predicted and desired output is almost perfect. The correlation coefficient (r) describes the difference between the desired output and actual output is shown in the performance table in Table2.

PERFORMANCE

DESIRED OUTPUT

MSE NMSE MAE Min Abs Error Max Abs Error R Percent Correct

0.209522132 1.164459543 0.23114814 9.90854E-07 1.015785003 0.498099362 73.07692308

ACTUAL NETWORK OUTPUT 0.212860733 1.18301446 0.25780224 6.66044E-06 0.998857054 0.498099362 75

Table 2: Performance Tablet

www.borjournals.com

Blue Ocean Research Journals

14

Journal of Engineering, Computers & Applied Sciences (JEC&AS) Volume 2, No.9, September 2013

4. Conclusion This work is aimed to find the success rate of IVF treatment using Artificial Neural Network (ANN). The ANN mostly shows accurate result in IVF success rate estimation. The data‟s (information) is collected from various hospitals and research centers are pre-processed and sent as input to the optimized Artificial Neural Network to train, cross validation, testing and predict the IVF success rate. The results of actual network output are compared with desired output. Finally desired output and actual network output are compared, and the result shows the most perfect output. This work shows 73% of accuracy in the results.

5. References [1] Adriana Albu , Loredana Ungureanu “Artiicial Neural Network in Medicine” International Journal of Computer science. [2] M. Durairaj, and K. Meena. “Application of Artificial Neural Network for Predicting Fertilization Potential of Frozen Spermatozoa of Cattle and Buffalo, “International Journal of Computer Science and System Analysis, pp. 1-10, 2008. [3] Greenwood, D. “An overview of neural networks”. BehavSci, 36:1-33. [4] Kaufmann, S.J., Eastaugh, J.L., Snowden, S., Smye, S.W and Sharma. “The Application of artificial neural networks in predicting the outcome of in-vitro fertilization. Human Reproduction” (1997). [5] Maui Elfaki Yahia and Roman Mahmod. “Hybrid expert system of rough set and neural network”, Malaysian Journal of Computers Science, (1999) Vol. 12 No. 1. [6] M. Durairaj and K. Meena, “Application of Artificial Neural Network for Predicting Fertilization Potential of Frozen Spermatozoa of Cattle and Buffalo”, International Journal of Computer Science and System Analysis, Serials Publications, Vol. 2, No. 1, Jan-Jun 2008, pp. 110. [7] K. Meena, M. Durairaj and K. R. Subramanian, “Machine Learning Techniques of Artificial Neural Network Modeling to Predict Fertility Rate of Sperm from the Outcome of IVF Functional Tests”, International Journal of Computer Science and Applications. [8] B. Larsson, H. Rodriguez-Martinez, “Can we use in vitro fertilization tests to predict semen fertility?”, Animal Reproduction Science, 60- 61, 2000, pp. 327-336. [9] B. Larsson, H. Rodriguez-Martinez, “Can we use in vitro fertilization tests to predict semen

www.borjournals.com

ISSN No: 2319-5606

fertility?”, Animal Reproduction Science, 60- 61, 2000, pp. 327-336. [10] Yaile Caballero, Rafael Bello, Yanitza Salgado and Maria M. Garcia, “A Method to Edit Training Set Based on Rough Sets,” International Journal of Computational Intelligence Research, vol. 3:3, pp. 219-229, 2007. [11] M. Durairaj, K. Meena “A Hybrid Prediction System Using Rough Sets and Artificial Neural Network” International Journal of Innovative Technology and Creative Engineering (ISSN: 2045-8711) Vol.1 No.7 July 2011. [12] M. T. Hogan and M. Menhaj, “Training feed forwards networks Back Propagation algorithm”, IEEE Trans. Neural Networks 5 (6), 1994. [13] Online Resource www.wikipedia.com

Acknowledgment The authors thank Janani Fertility Centre, Trichy, Devi Hospital, Perambalur, Sri Ramakrishna Hospital, Coimbatore, Ishwarya Fertility Centre, Salem, R.J. Clinic, Coimbatore, Smile Hospital, Mettupalayam, Abi Polyclinic & Fertility Centre, Salem, Abirami Fertility Centre and Good luck Clinic, Coimbatore and for providing data and valuable technical inputs for this research work.

Blue Ocean Research Journals

15