|
Activity Number:
|
221
- Contributed Poster Presentations: Section on Statistics in Imaging
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Section on Statistics in Imaging
|
|
Abstract #313286
|
|
|
Title:
|
An Exploration of the Use of Statistical Machine Learning Techniques to Shorten the Duration of Dynamic PET-FDG Imaging Protocols in a Clinical Context
|
|
Author(s):
|
Qi Wu* and Finbarr O'Sullivan and Mark Muzi and David Mankoff
|
|
Companies:
|
University College Cork and University College Cork and University of Washington and University of Pennsylvania
|
|
Keywords:
|
PET Scans;
Metabolic Profiling;
Machine Learning
|
|
Abstract:
|
Positron emission tomography (PET) scanning with fluoro-deoxyglucose (FDG) has an important role in cancer diagnosis and treatment planning. The standard clinical protocol is focused on the recovery of FDG retention in tissue. Imaging is typically for 15 minutes about one hour after tracer injection. The next generation of Total-body scanners will give a unique potential for dynamic acquisitions that could yield more detailed information about FDG delivery and metabolism. For this to be clinically viable, protocols would need to be of short duration and still be able to provide reliable estimates of late-time FDG retention. We consider a series of PET-FDG studies in brain and breast cancer patients where 60-90 minutes of dynamic scanning was carried out. By re-sampling the dynamic scanning data were able to study the ability to evaluate/predict late FDG retention from early dynamic scans information. A metabolic profiling procedure is used to extract feature vectors from the raw dynamic data. Several techniques including Multiple Linear Regression, Generalized Additive Models, Random Forests and Neural Networks were applied. The results are found to be quite promising.
|
Authors who are presenting talks have a * after their name.
