Brain multi modality image inpainting via deep learning based edge region generative adversarial network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Brain tumors (BT) are critical neurological diseases impacting central nervous system functions.
• Analyzing diseased brain images is challenging due to tissue abnormalities affecting image processing.
• Existing methods struggle with accurate tissue segmentation and non-rigid registration in brain MRI.

Purpose of the Study:

• To propose a Deep-Supervised Generative Adversarial Network (DS-GAN) model for effective inpainting of brain MRI images.
• To enhance the accuracy and reliability of brain tumor image analysis.
• To overcome limitations of current techniques in processing abnormal brain tissues.

Main Methods:

• Image segmentation using a Gated Shape Convolutional Neural Network (GS-CNN).
• Development of an Edge Reconstruction Generative Adversarial Network (EGAN) for generating missing image edges.
• Implementation of the DS-GAN model for brain MRI inpainting.

Main Results:

• The DS-GAN model achieved high performance metrics: Jaccard Index (JI) of 0.82 and Dice Index (DI) of 0.86.
• Quantitative results for DS-GAN: L1 loss 2.18, PSNR 0.972, SSIM 32.04, and MSE 26.42.
• The proposed DS-GAN model demonstrated an overall accuracy of 99.18%, surpassing existing techniques.

Conclusions:

• The DS-GAN model offers a significant advancement in brain MRI inpainting.
• Accurate inpainting of brain MRI images is crucial for improved brain tumor diagnosis and analysis.
• The proposed method effectively addresses challenges posed by abnormal brain tissues in image processing.