Comparison of three training algorithms of Artificial Neural Networks in kidney stone detection from CT scan images

Chronic kidney disease (CKD) is a invisible disease. Accurate prediction of CKD progression at the right time is necessary to reduce its costs and mortality. In the present study, we investigated three artificial neural network training algorithms in the early detection of kidney stones from CT scan images. The images obtained from CT-scan have noise and spots and have low quality, therefore pre-processing is the first important step. After the pre-processing stage, the features were extracted using the gray level co-occurrence matrix and then the optimal features were selected. The obtained features were considered as input and labels as targets and finally we determined the output with specific data. The obtained results showed that The training algorithms of Bayesian regularization, Levenberg-Marquart, and scaled conjugate gradient had the recognition accuracy of ۹۹/۹۶%, ۹۸/۷۸%, and ۹۳/۳۳%, respectively.