Enhancement of Perivascular Spaces Using Densely Connected Deep Convolutional Neural Network

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Perivascular spaces (PVS) are implicated in various brain diseases.
• Quantifying PVS is challenging due to their indistinct appearance in MR images.

Purpose of the Study:

• To develop a deep learning method for enhancing Magnetic Resonance (MR) images to improve PVS visualization.
• To introduce a novel, very deep 3D convolutional neural network for accurate PVS enhancement.

Main Methods:

• A very deep 3D convolutional neural network with dense connections and skip connections was proposed.
• The network leverages rich contextual information from multi-level features to mitigate gradient vanishing.
• The method was evaluated using 17 7T MR images with twofold cross-validation.

Main Results:

• The proposed deep learning method significantly enhances PVS visualization in MR images.
• The network effectively utilizes contextual information for accurate image prediction.
• Experimental results demonstrate superior performance compared to previous PVS enhancement methods.

Conclusions:

• The developed deep learning approach offers a more effective way to visualize PVS in brain MR images.
• This method has the potential to improve the diagnosis and study of PVS-related brain diseases.
• The proposed 3D convolutional neural network architecture is highly effective for medical image enhancement.