An edge-directed interpolation method for fetal spine MR images

Area of Science:

• Medical Imaging
• Image Processing
• Prenatal Diagnostics

Background:

• Fetal spinal magnetic resonance imaging (MRI) is vital for assessing fetal development, particularly for suspected spinal malformations when ultrasound is insufficient.
• Low resolution in fetal spine MRI limits diagnostic accuracy; high-resolution images are needed but current interpolation methods often fail to preserve critical edge structures.
• Preserving edge structures is essential for accurate detection, classification, and diagnosis of fetal spinal malformations.

Purpose of the Study:

• To develop and evaluate a novel edge-directed interpolation (EDI) method for enhancing the resolution of fetal spine MRI images.
• To assess the method's ability to preserve structural integrity, specifically edge details, during the interpolation process.
• To compare the performance of the proposed EDI method against existing techniques using quantitative metrics and visual analysis.

Main Methods:

• Proposed an edge-directed interpolation (EDI) method utilizing Canny edge detection to guide pixel modification for enhancing fetal spine MR images.
• Employed a bilinear interpolation as an initial step, followed by a dual strategy to refine edge structures based on edge information from low-resolution (LR) and high-resolution (HR) images.
• Validated the method by comparing generated HR images with those from four other EDI techniques, using six quantitative metrics and subjective analysis of regions of interest (ROI).

Main Results:

• The proposed EDI method generated high-resolution (HR) fetal spine MR images with enhanced details and superior preservation of edge structures compared to other methods.
• Quantitative evaluation showed the proposed method outperformed others in signal-to-noise ratio (SNR), peak signal-to-noise ratio (PSNR), structure similarity index (SSIM), feature similarity index (FSIM), and mutual information (MI).
• Visual analysis confirmed better consistency of edge structures with original images, achieved within seconds of computation time.

Conclusions:

• The developed EDI method effectively classifies edge orientations and preserves structural details, producing high-quality HR fetal spine MR images suitable for real-time clinical applications.
• The method demonstrates robustness against noise and artifacts, offering a significant improvement over other tested interpolation techniques.
• The proposed approach provides a valuable tool for improving the accuracy and efficiency of prenatal diagnosis of fetal spinal conditions.