Machine Learning in PET: from Photon Detection to Quantitative Image Reconstruction

Area of Science:

• Nuclear Medicine
• Medical Imaging
• Artificial Intelligence

Background:

• Traditional nuclear medicine imaging relies on basic signal processing for detector data.
• Advancements in waveform digitizers enable sophisticated analysis of high-energy photon signals.
• Existing methods for image reconstruction and correction can be computationally intensive.

Purpose of the Study:

• To review the applications of machine learning in nuclear medicine.
• To highlight ML's role in photon detection and quantitative image reconstruction.
• To discuss ML's impact on improving accuracy and efficiency in nuclear imaging.

Main Methods:

• Application of machine learning algorithms to detector signal processing.
• Utilizing ML for estimating position and arrival time of high-energy photons.
• Employing ML techniques for quantitative image reconstruction, including correction factors and noise reduction.

Main Results:

• ML accurately estimates photon position and arrival time, advancing detector capabilities.
• ML-based methods provide faster alternatives for scatter and attenuation correction.
• AI-driven approaches enable data-driven mapping for complex functions like PET/MR attenuation.

Conclusions:

• Machine learning is revolutionizing nuclear medicine by enhancing detector performance and image reconstruction.
• ML offers significant improvements in speed and accuracy for critical imaging tasks.
• The integration of ML paves the way for more efficient and precise nuclear imaging diagnostics.