Prediction the success rate of intracytoplasmic sperm injection (ICSI) using artificial neural networks and logistic regression

Intra-cytoplasmic sperm injection (ICSI) in infertile couples has created a great improvement for treatment of these patients. Unfortunately, despite of high cost of doing ICSI the rate of success is not acceptable and failing pregnancy put a heavy anxiety to couples. This study is aimed to make an effort in order to extract the best predictors for predict the success rate of intra cytoplasmic sperm injection and promote accuracy, sensivity and specificity by the use of artificial neural network and logistic regression.

345 patients received ICSI treatment and each of which constructed of 54 numerical and nominal records. This database was randomly divided into the estimation (n=276) and validation (n=69) data set. The models were used based on binary logistic regression (BLR) feature selection tools and Levenberg-Marquardth neural network classifier. Finally, models were evaluated using important criteria such as accuracy, sensitivity and specificity.

The best output of the B LR model by using 54 variables revealed accuracy (97%), sensitivity (86%)and specificity (%94). The best output of the LMNN model using Reduced dataset consisted of n=29 with a feature vector side yielded the accuracy (82%) and sensitivity (%92) and specificity (%76).

Our result demonstrated that BLR model outperformed highlighting the great power of BLR in success rate of ICSI prediction while using binary output.