Quantitative Molecular Positron Emission Tomography Imaging Using Advanced Deep Learning Techniques.

Area of Science:

• Molecular imaging research
• Artificial intelligence in medicine
• Quantitative imaging techniques

Background:

• High-performance computing and AI, particularly machine learning (ML) and deep learning (DL), are driving innovation in molecular imaging.
• AI applications span PET instrumentation, image reconstruction, analysis, and computer-aided diagnosis.
• There's a growing interest in quantitative molecular imaging using ML/DL over the past decade.

Purpose of the Study:

• To review the principles and applications of ML/DL in quantitative molecular imaging.
• To discuss the use of ML/DL in obtaining accurate PET data, including denoising and correction algorithms.
• To explore the role of ML/DL in quantifying tracer uptake and tumor volume for treatment monitoring and response prediction.

Main Methods:

• Review of literature on AI, ML/DL applications in molecular imaging.
• Discussion of algorithms for image reconstruction, denoising, and quantification.
• Analysis of ML/DL's role in treatment planning and response prediction.

Main Results:

• AI/ML/DL techniques are widely applied in molecular imaging, from instrument design to data analysis.
• ML/DL algorithms improve quantitative accuracy of PET data by addressing physical degradations.
• These techniques are crucial for monitoring treatment response and planning radiation therapy.

Conclusions:

• ML/DL techniques are transforming quantitative molecular imaging, offering advanced capabilities for diagnosis and treatment.
• Challenges and future opportunities exist for adopting ML/DL in multimodality imaging.
• Accurate quantitative PET data acquisition and analysis are significantly enhanced by AI.