Program Information
A Fully Automated CT Tool to Facilitate Phantom Image QA for Quantitative Imaging in Clinical Trials
M Wahi-Anwar*, P Lo , H Kim , M Brown , M McNitt-Gray , UCLA Radiological Sciences, Los Angeles, CA
Presentations
SU-G-206-1 (Sunday, July 31, 2016) 4:00 PM - 6:00 PM Room: 206
Purpose: The use of Quantitative Imaging (QI) methods in Clinical Trials requires both verification of adherence to a specified protocol and an assessment of scanner performance under that protocol, which are currently accomplished manually. This work introduces automated phantom identification and image QA measure extraction towards a fully-automated CT phantom QA system to perform these functions and facilitate the use of Quantitative Imaging methods in clinical trials.
Methods: This study used a retrospective cohort of CT phantom scans from existing clinical trial protocols - totaling 84 phantoms, across 3 phantom types using various scanners and protocols.
The QA system identifies the input phantom scan through an ensemble of threshold-based classifiers. Each classifier - corresponding to a phantom type - contains a template slice, which is compared to the input scan on a slice-by-slice basis, resulting in slice-wise similarity metric values for each slice compared. Pre-trained thresholds (established from a training set of phantom images matching the template type) are used to filter the similarity distribution, and the slice with the most optimal local mean similarity, with local neighboring slices meeting the threshold requirement, is chosen as the classifier’s matched slice (if it existed). The classifier with the matched slice possessing the most optimal local mean similarity is then chosen as the ensemble’s best matching slice. If the best matching slice exists, image QA algorithm and ROIs corresponding to the matching classifier extracted the image QA measures.
Results: Automated phantom identification performed with 84.5% accuracy and 88.8% sensitivity on 84 phantoms. Automated image quality measurements (following standard protocol) on identified water phantoms (n=35) matched user QA decisions with 100% accuracy.
Conclusion: We provide a fully-automated CT phantom QA system consistent with manual QA performance. Further work will include parallel component to automatically verify image acquisition parameters and automated adherence to specifications.
Funding Support, Disclosures, and Conflict of Interest: Institutional research agreement, Siemens Healthcare Past recipient, research grant support, Siemens Healthcare Consultant, Toshiba America Medical Systems Consultant, Samsung Electronics NIH Grant support from: U01 CA181156
Contact Email: