سالار بابازاده بهستانی

In this thesis, the ۳D MRI image segmentation methods have been investigated and improved using bayesian neural networks and variable self-encoder networks. The main goal of this research is to increase accuracy and efficiency in the diagnosis and analysis of medical images, especially in the field of brain tumors. Considering the complexity and high volume of medical data, the use of advanced deep learning methods and Bayesian neural networks and variable self-encoder can lead to a significant improvement in the image segmentation and analysis process.

In this research, firstly, theoretical foundations related to ۳D MRI images and segmentation of ۳D images and statistical deep learning networks for segmentation of these images have been introduced. In the following, the preprocessing of ۳D images and especially the introduction of Bayesian neural networks and variable self-encoder and their applications in medical image segmentation have been discussed.

The results of this research show that the use of variable self-encoding neural networks can improve accuracy and reliability in the diagnosis and analysis of medical images compared to Bayesian convolutional neural networks and also reduce errors caused by noise and changes. that the value of the Dice coefficient for the variable self-encoding neural network for the three tumor regions, tumor core and tumor region is ۰.۹۱, ۰.۸۶ and ۰.۸۲ for the validation data and ۰.۸۸, ۰.۸۱ and ۰.۷۶ for the validation data respectively.

Finally, suggestions for future research and improvement of existing methods are provided. This research can be considered as an important step in improving the quality of medical services and improving the health of patients.