A Research for Segmentation of Brain Tumors Based on GAN Model

Linh Khanh Phung, Sinh Van Nguyen, Tan Duy Le, Marcin Maleszka

Abstract: Analysis of medical image is a useful method that can support doctors in medical diagnosis. The development of deep learning models is essential and widely applied in image processing and computer vision. Application of machine learning and artificial intelligent in brain tumor diagnosis brings an accuracy and efficiency in medical treatment field. In this research paper, we present a method for determining and segmenting the brain tumor region in the medical image dataset based on 3D Generative Adversarial Network (3D-GAN) model. We first explore the state-of-the-art methods and recent approaches in such field. Our proposed 3D-GAN model consist of three steps: (i) pre-processing data, (ii) building an architecture of multi-scaled GAN model, and (iii) modifying loss function. The last our contribution is creating an application to visualize 3D models that representation of medical resonance brain images with the incorporation of the chosen models to determine exactly the region containing brain tumors. Comparing to the existing methods, our proposed model obtained better performance and accuracy.

External IDs:dblp:conf/aciids/PhungNLM22