Transformers in medical imaging: A survey

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Computer Vision

Background:

• Transformers have achieved state-of-the-art results in natural language processing and computer vision.
• Their ability to capture global context is advantageous over the local receptive fields of Convolutional Neural Networks (CNNs).
• This has led to increasing interest in applying Transformers to medical imaging tasks.

Purpose of the Study:

• To provide a comprehensive review of Transformer applications in medical imaging.
• To cover architectural designs, specific tasks, challenges, and future directions.
• To serve as an up-to-date reference for researchers in the field.

Main Methods:

• Surveying existing literature on Transformers in medical imaging.
• Categorizing applications including segmentation, detection, classification, restoration, synthesis, registration, and report generation.
• Analyzing challenges, trends, and future research avenues.

Main Results:

• Transformers are successfully applied across various medical imaging tasks, demonstrating significant potential.
• Key challenges and trends specific to each application are identified.
• A critical discussion of the field's current state and future outlook is provided.

Conclusions:

• Transformers represent a significant advancement in medical imaging analysis.
• Addressing identified challenges will further unlock their potential.
• This survey provides a roadmap for future research and development.