Learning image-based spatial transformations via convolutional neural networks: A review.

Area of Science:

• Medical image analysis
• Artificial Intelligence
• Quantitative imaging

Background:

• Deep learning (DL) and advanced hardware have transformed quantitative medical image analysis.
• Convolutional neural networks (CNNs) offer generalizability, efficiency, and open-source accessibility, driving paradigm shifts.
• The impact of DL is evident in literature, conferences, and competitions within medical imaging.

Purpose of the Study:

• To review state-of-the-art deep learning approaches for image registration and transformation optimization.
• To contextualize advancements in deep learning for image registration within the broader DL field.
• To encourage further development and adoption of DL techniques in medical imaging research.

Main Methods:

• Review of state-of-the-art deep learning methods, focusing on CNNs.
• Analysis of learning, prediction, and optimization of image transformations.
• Contextualization of image registration advancements within the deep learning landscape.

Main Results:

• Deep learning, particularly CNNs, has significantly impacted quantitative medical image analysis.
• Image registration using DL shows substantial progress, though less explored than segmentation.
• Significant advancements have been presented across various research venues.

Conclusions:

• Deep learning techniques are increasingly vital for medical image analysis.
• Further research and application of DL in medical image registration are encouraged.
• This review aims to facilitate the leveraging of DL for medical imaging research.