Low-Rank Based Image Analyses for Pathological MR Image Segmentation and Recovery

Area of Science:

• Medical Imaging
• Computational Biology
• Machine Learning

Background:

• Pathologies in MR brain images complicate analysis tasks like segmentation and registration.
• Existing low-rank and sparse decomposition (LSD) methods struggle with reliable image recovery due to a lack of constraint between components.

Purpose of the Study:

• To develop a novel method for simultaneous recovery and segmentation of pathological MR brain images.
• To address the limitations of conventional LSD approaches in producing reliable recovered images.

Main Methods:

• Propose a transformed low-rank and structured sparse decomposition (TLS2D) method.
• Integrate structured sparse constraint, LSD, and image alignment into a unified scheme.
• Utilize combined structured sparse and image saliency as adaptive sparsity constraint for image recovery.

Main Results:

• The TLS2D method demonstrates robustness in distinguishing pathological regions.
• Experimental results on synthetic and real MR brain tumor images show effective image recovery.
• The method achieves satisfactory tumor segmentation accuracy.

Conclusions:

• The proposed TLS2D method offers an effective solution for simultaneous pathological MR brain image recovery and segmentation.
• TLS2D overcomes limitations of conventional LSD by integrating structural constraints and image alignment.
• The approach shows significant potential for clinical applications in neuroimaging analysis.