An Automatic Parameter Decision System of Bilateral Filtering with GPU-Based Acceleration for Brain MR Images.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Bilateral filters are crucial for image denoising in medical applications like segmentation and classification.
• Manual adjustment of bilateral filter parameters is time-consuming and image-dependent.
• Existing methods often lack efficiency and accuracy in automated denoising.

Purpose of the Study:

• To develop a computer-aided system for automatic bilateral filter parameter selection.
• To enhance the efficiency and accuracy of brain MRI denoising using neural networks and GPU acceleration.
• To provide a parameter-free solution for medical image restoration.

Main Methods:

• Developed a GPU-accelerated bilateral filter for faster computation.
• Utilized a back propagation network (BPN) with image texture features to predict filter parameters.
• Employed k-fold cross-validation for performance evaluation on T1-weighted brain MR images.

Main Results:

• Achieved a 208x speed-up compared to CPU-based computation.
• The parameter prediction system demonstrated high accuracy with a Mean Absolute Percentage Error (MAPE) of 6%.
• Outperformed state-of-the-art methods in noise removal, as measured by Peak Signal-to-Noise Ratio (PSNR).

Conclusions:

• The proposed system automates brain MR image denoising effectively using texture features and BPN.
• GPU-based bilateral filtering significantly accelerates the denoising process.
• This automatic restoration framework offers advantages for various medical image processing tasks.