Turkish Journal of Gastroenterology
Original Article

The rate of mucosal healing by azathioprine therapy and prediction by artificial systems

1.

Department of Electrical and Electronics Engineering, Gazi University Faculty of Engineering, Ankara, Turkey

2.

Department of Gastroenterology, Türkiye Yüksek İhtisas University Hospital, Ankara, Turkey

3.

Department of Gastroenterology, Bezmialem Vakıf University Faculty of Medicine, İstanbul, Turkey

4.

Department of Gastroenterology, Ankara University Faculty of Medicine, Ankara, Turkey

Turk J Gastroenterol 2015; 26: 315-321
DOI: 10.5152/tjg.2015.0199
Read: 439 Downloads: 93 Published: 25 July 2019

Abstract

Background/Aims: We aimed to assess the effect of azathioprine on mucosal healing in patients with inflammatory bowel diseases (IBD). Artificial neural networks were applied to IBD data for predicting mucosal remission.

 

Materials and Methods: Two thousand seven hundred patients with IBD were evaluated. According to the computer-based study, data of 129 patients with IBD were used. Artificial neural networks were performed and tested.

 

Results: Endoscopic mucosal healing was found in 37% patients with IBD. Male gender group showed a negative impact on the efficacy of azathioprine (p<0.05). Responder patients with IBD were older than the nonresponder (p<0.05) patients. According to this study, the cascade-forward neural network study provides 79.1% correct results. In addition to a 0.16033 training error, mean square error (MSE) was taken at the 16th epoch from the feed-forward back-propagation neural network. This neural structure, used for predicting mucosal remission with azathioprine, was also validated.

 

 

Conclusion: Analyzing all parameters within each other to azathioprine therapy were shown that which parameters gave better healing were determined by statistical, and for the most weighted six input parameters, artificial neural network structures were constructed. In this study, feed-forward back-propagation and cascade-forward artificial neural network models were used.

Files
EISSN 2148-5607
3