Deep Learning and Its Application to Function Approximation for MR in Medicine: An Overview.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning (DL) is increasingly impactful in medical sciences, particularly magnetic resonance (MR).
• Understanding DL's technical aspects can be challenging for those outside machine learning (ML).

Purpose of the Study:

• To provide an accessible overview of DL and its applications in medical MR.
• To enable readers to develop and understand DL applications in MR.

Main Methods:

• Explanation of core AI, ML, and DL concepts, including deep neural networks (DNNs) and layers.
• Coverage of essential DL technologies: loss functions, optimization, regularization, and network architectures.
• Overview of DL applications in MR, including RF pulse design, reconstruction, and image segmentation.

Main Results:

• DL provides a framework for function approximation in MR.
• DNNs, composed of layers, are central to DL's capabilities.
• Numerous MR applications benefit from DL, from pulse sequence optimization to image synthesis.

Conclusions:

• DL offers significant potential for advancing MR applications.
• This article demystifies DL for a broader audience in medical MR.
• Familiarity with DL principles can drive innovation in medical imaging.