Deep learning: definition and perspectives for thoracic imaging.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Machine learning, especially deep learning and convolutional neural networks (CNNs), is increasingly relevant in clinical practice.
• CNNs have demonstrated superior performance in various clinical scenarios compared to classical machine learning algorithms and human experts.
• Thoracic imaging, particularly chest radiography, is a prime area for deep learning due to abundant data.

Purpose of the Study:

• To define deep learning methods and their applications in thoracic imaging.
• To review the current state and future potential of deep learning in radiology.
• To provide an overview of convolutional neural networks (CNNs) in medical imaging.

Main Methods:

• Review of current literature on deep learning applications in thoracic imaging.
• Focus on convolutional neural networks (CNNs) as a key deep learning architecture.
• Discussion of algorithm development for automated reporting in chest radiography and CT imaging.

Main Results:

• Deep learning techniques, especially CNNs, outperform traditional machine learning in many radiological tasks.
• CNNs are the most prevalent deep learning architecture in medical imaging.
• Numerous deep learning algorithms are under development, with potential for near-future clinical integration.

Conclusions:

• Deep learning methods, particularly CNNs, are rapidly advancing and showing significant potential in thoracic imaging.
• These AI tools are expected to complement and enhance the capabilities of radiologists.
• The integration of deep learning into clinical practice is likely to increase, improving diagnostic accuracy and efficiency.