Low contrast lesion detection in PET using the Bootstrap method
2015
In a low contrast lesion detection experiment, large amounts of repeated measurements on the same object are typically required in order to achieve a good estimation of statistics. One way to obtain these many realizations can be through repeatedly scanning the low contrast object, which however, is a challenge with Positron Emission Tomography (PET) imaging due to the decaying nature of radionuclides using radiopharmaceuticals like Fluorodeoxyglucose (FDG). In order to calculate lesion detectability from a limited number PET FDG scans, the bootstrap resampling method was used to create n realizations. That is, from the same sinogram data, we randomly draw n bootstrap samples of size k with replacement for prompt and random coincidence events. To validate the proposed method, 60 parallel scans (3 min each) of a NEMA phantom filled with FDG were obtained by using listmode data from a gated acquisition on a PET/CT scanner. We resampled the sinogram of full data into 60 replicates with each having the same total number of prompts and random events as the replayed scan data. PET images were reconstructed from both scan data and bootstrap data using Time-of-Flight (TOF) and non-TOF Ordered-Subset Expectation Maximization (OSEM) algorithms. A 10-channel dense Difference-of-Gaussian (DoG) Channelized-Hotelling Observer (CHO) was developed with internal noise added to channel outputs and applied to the PET images. CHO showed close detectability on the 4 hot spheres (10mm, 13mm, 17mm and 22mm) of the NEMA phantom between scan data and bootstrap data, on both TOF and non-TOF PET images. We applied the bootstrap method to another scan of NEMA phantom with fixed contrast 4:1 and smaller spheres (5mm, 6mm, 8mm and 10mm), and created 150 realizations (10 min each). We conclude the bootstrap method is useful for low contrast lesion detection study using CHO.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
6
References
0
Citations
NaN
KQI