GPU-Accelerated GLRLM Algorithm for Feature Extraction of MRI.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Analysis

Background:

• Gray Level Run Length Matrix (GLRLM) is crucial for extracting statistical features from medical images like MRIs.
• These features are vital for texture pattern identification in artificial neural networks.
• Current GLRLM computation is slow for high-resolution images or multiple Regions of Interest (ROIs).

Purpose of the Study:

• To accelerate the computationally intensive GLRLM construction and feature extraction process.
• To develop a new paradigm for simultaneous GLRLM generation and feature extraction for multiple ROIs.
• To leverage Graphics Processing Unit (GPU) computing for enhanced medical image preprocessing.

Main Methods:

• Proposed a novel paradigm utilizing parallel primitives for GLRLM generation.
• Implemented simultaneous feature extraction for multiple ROIs within a single image.
• Employed Graphics Processing Unit (GPU) computing for parallel processing.

Main Results:

• The proposed paradigm significantly accelerates GLRLM and feature extraction.
• Achieved over a 5-fold speed increase compared to optimized serial methods.
• Demonstrated ease of implementation for the parallel processing paradigm.

Conclusions:

• The GPU-accelerated paradigm offers a substantial speedup for medical image texture analysis.
• This approach effectively addresses the computational challenges of processing large or complex medical images.
• Facilitates more efficient preprocessing for machine learning applications in medical imaging.