---

Background and purpose: Since the human health is an essential issue in medical sciences, accurate predicting the individual's disease status is of great importance. Therefore, predicting with models minimum error and maximum certainty should be used. This study used artificial neural network model for predicting coronary artery disease (CAD) because it is more precise Comared to after models. Materials and methods: Multilayer perceptron (MLP) with post propagation error algorithm (EBP) for assessing the coronary artery disease was implemented on 150 patients admitted to the Mazandaran Heart Center, Sari. Then, based on the 80% of the available data, an artificial neural network with NN (14, 12, 1), sigmoid transfer function and 1500 epochs were designed and trained. The data were fed into Excel program and then softwares for artificial neural network designing such as Pythia-Neural Network were employed. Results: Mean square of the error in training step was decreased to the level of 0.0238 and sensitivity and specificity rates obtained were 0.96 and 1. In the end, the model correctly categorized some healthy individuals who did not require angiography and the treatment related to coronary artery diseases. Conclusion: Due to the high specificity index, this model prevents side effects of angiography in patients who do not require such interventions. Moreover, due to high sensitivity, it can diagnose the patients who really need such diagnostic measures.

---

Background and purpose: Some diseases such as osteoporosis may have no symptom but suddenly cause fractures in different parts of body such as spine, chest, hands and legs, thereby resulting in very painful death in old people. According to a report by Iran’s ministry of health 4.6% of people aged 20 to 70 years in Iran are affected by osteoporosis in the spine. This study aimed at determining the factors influencing the incidence of osteoporosis and also providing a predictive model to speed up the detection and reduce diagnostic costs. Material and Methods: Data was collected by interviewing 670 patients in an orthopedic clinic. The information included demographic information, lifestyle and diseases, and the results of DEXA scan. In this paper, a new model based on the standard methodology CRISP is presented. In modeling, three known data mining methods, the CHAID, C5.0 decision tree, and neural network were used. For data analysis Celementine V.12.0 was used.. Results: In this study, for the first time in Iran, the characteristics affecting osteoporosis in patients has been studied. Using data mining techniques, influencing characteristics of the disease have been identified. According to the created decision tree, some rules are derived that can be used as a model for the prediction of patient’s status. Accuracy of built models using the algorithms C.5.0, CHAID, and neural networks were compared. Each algorithm was observed to act better in predicting osteoporosis in a specified group of people. Conclusion: The accuracy of artificial neural networks algorithm is higher than that of the decision tree algorithm. In this study the most affected factors on osteoporosis were detected. According to the created rules for a new instant with specified features, we can predict whether a patient will probably suffer from osteoporosis or not.

---

Background and purpose: Intelligent methods such as artificial neural networks (ANN) have been recently used as an efficient model for prediction and classification of tumors. Diagnosis of benign and malignant breast tumors based on morphological, clinical and demographic features without using invasive paraclinical methods is very important. The aim of this study was to provide a neural network model to predict the status of breast tumors and compare its efficacy with the common regression model.

Materials and methods: In this study, Wisconsin breast cancer database was used. It was obtained from cytology results of the breast tumors of 683 patients. In the proposed model different features such as clump thickness, uniformity of cell size, uniformity of cell shape, etc. were used as input variables. We applied the genetic algorithm (GA) for determination of the best structure and training of multi-layer NN model was implemented in MATLAB. The performance of proposed NN model was compared appling logistic regression (LR) in SPSS. 5-fold cross validation was used for accurate calculation of the performance of the models.

Results: The results found GA capable of determining the best structure for a multi-layer NN and train it properly. In different performances the best NN structure was NN(9-8-6-1) with an average accuracy, sensitivity, specificity, and AUC (area under ROC curve) of 0.971, 0.988, 0.962, and 0.9955, respectively, while the values of the corresponding parameters for LR were 0.968, 0.975, 0.964 and 0.9954, respectively.

Conclusion: The achieved ANN model could be used as a method with high sensitivity and specificity alongside common non-invasive diagnostic methods as a diagnosis support system to identify benign and malignant breast tumors.

---

Background and purpose: Despite rapid progress in medical treatments and acute care technology during the past 30 years alongside increasing costs of medical care, the analysis of outcomes such as mortality risk have been a challenge in intensive care units. The purpose of this study was to predict the mortality rate of premature infants in neonatal intensive care unit (NICU) using artificial neural network model.

Materials and methods: This study was performed using the medical records of 100 preterm infants (less than 37 weeks gestation) in Mashhad Qaem Hospital, Iran, during 2007-2010 applying MATLAB. Twenty one variables were used of which 80% were for artificial neural network training and 20 percent were for testing the designed model. To prevent the dispersion of information we used information classification code system and the codes were used to design and test the artificial neural network model.

Results: Per 60 neurons and 20 replication optimum validity was obtained (95.2% in training and 94.56% in experimental stage). The replications were not continued more, since in this case the algorithm would have gone towards overtraining.

Conclusion: This study introduced a method for establishing ANN models in estimating the probability of mortality in premature infants using 21 variables. This model may be used for prediction of many other consequences in NICU such as mechanical ventilation duration and complications such as abnormalities in neuroimaging, necrotizing enterocolitis and bronchopulmonary dysplasia.

---

Background and purpose: Nonlinear analysis methods for quantitative structure–activity relationship (QSAR) studies better describe molecular behaviors, than linear analysis. Artificial neural networks are mathematical models and algorithms which imitate the information process and learning of human brain. Some S-alkyl derivatives of thiosemicarbazone are shown to be beneficial in prevention and treatment of mycobacterial infections and this study seeks to find out the relationship between structural features and the anti-tuberculosis activity of these compounds.
Materials and methods: Multiple linear regression and Bayesian regularized artificial neural network (BRANN) for 47 compounds of thiosemicarbazone derivatives were designed using QSAR approaches. Descriptors were selected from a pool of 343 descriptors by stepwise selection and backward elimination. A three layer Bayesian regularized back-propagation feed-forward network was designed, optimized, and evaluated using MATLAB version R2009a.
Results: The best model with 6 descriptors was found using multiple linear regression analysis: Log MIC= 2.592 + (0.067 ± 0.018) PMIX – (0.066 ± 0.017) PMIZ – (1.706 ± 1.600) Qneg – (0.235 ± 0.039) RDF030p + (0.118 ± 0.026) RDF 140u – (0.064 ± 0.021) RDF060p. The best BRANN model was a three-layer network with three nodes in its hidden layer.
Conclusion: The BRANN model has a better predictive power than linear models and may better predict the anti-tuberculosis activity of new compounds with similar backbone of thiosemicarbazone moiety.