Development of a deep learning method for phase retrieval image enhancement in phase contrast microcomputed...

Area of Science:

• Medical Imaging
• Materials Science
• Artificial Intelligence

Background:

• Conventional absorption-based microcomputed tomography (µCT) struggles to visualize low-density materials like hydrogels.
• X-ray phase contrast imaging, specifically propagation-based imaging with microcomputed tomography (PBI-µCT), offers improved visualization but faces challenges with noise and quantitative accuracy.
• Existing phase retrieval (PR) algorithms can improve signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) but often lead to over-smoothing and inaccuracies.

Purpose of the Study:

• To develop a novel deep learning-based method, Edge View Enhanced Phase Retrieval (EVEPR), to enhance PBI-µCT image quality for low-density materials.
• To improve the segmentation accuracy and efficiency of hydrogel constructs in vitro and ex vivo.
• To overcome the limitations of conventional PR algorithms, such as over-smoothing and noise susceptibility.

Main Methods:

• Developed EVEPR by integrating denoised edge-enhanced contrast (EEC) and phase retrieval (PR) images.
• Trained a deep convolutional neural network (CNN) using paired denoised EEC and PR images on a dataset-to-dataset basis.
• The CNN learned to combine high-frequency details from EEC images with area contrast from PR images.

Main Results:

• EVEPR demonstrated enhanced area contrast beyond conventional PR methods, significantly improving SNR and CNR.
• The enhanced CNR facilitated more efficient and accurate segmentation of low-density hydrogel constructs.
• Applied to in vitro and ex vivo PBI-µCT images, EVEPR provided superior visibility and consistency of hydrogel constructs, reducing manual segmentation adjustments.

Conclusions:

• EVEPR is a robust post-image processing method that significantly enhances PBI-µCT image quality for low-density materials.
• The method effectively addresses over-smoothing and noise issues inherent in conventional PBI-µCT processing.
• EVEPR enables efficient and accurate in vitro and ex vivo image processing and segmentation, facilitating the creation of large datasets for data-driven applications.