Coincidence events classification in PET tomography using machine learning methods
The source of the image in positron emission tomography (PET) is the detection of pairs of gamma photons produced by the annihilation of a positron with an electron. However, the recorded coincidence events include not only true signals, but also noise: Compton-scattered events (in the patient's body and in the detector crystals) and random events. Their effective identification can significantly improve image reconstruction quality and measurement accuracy. The presentation will discuss the application of deep learning to the classification of coincidence events. The model, based on a 3D CNN architecture, processes a spatial map extended with physical characteristics of coincidences, such as photon energies, detection time differences, and event geometry. The current results and prospects for the development of the method will be discussed.
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
- Nowowiejska 15/19
- Warsaw, Mazowieckie
- Poland 00-665
- Building: Faculty of Electronics and Information Technology
- Room Number: 229
Speakers
Michał