A convolutional neural network based approach for generating full PET/CT image series from shorter scan time

2019 Fiscal Year Research-status Report

• Project/Area Number: 19K20685
• Research Institution: National Institutes for Quantum and Radiological Science and Technology
• Principal Investigator: BhusalChhatkuli Ritu 国立研究開発法人量子科学技術研究開発機構, 放射線医学総合研究所 分子イメージング診断治療研究部, 研究員(任非) (50836591)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: PET/CT / neural networks / list mode PET / image generation

Outline of Annual Research Achievements

- PET/CT data from 13 Pancreatic Cancer patients have been collected.
- Those collected data have been used for analysis using convolutional neural networks.
- The network was trained using the dynamic list mode PET/CT scans as training images and the delayed scan as the label/target image.
- Leave one out cross validation technique was used for the validation.
- The generated delayed images were analysed to evaluate the performance of the convolutional neural network.
- Analysis shows successful delayed scan generation with minimum error.

Current Status of Research Progress

2: Research has progressed on the whole more than it was originally planned.

Reason

- 13 patient datasets has already been collected.
- Though further data sets required, the result was significant enough to suggest the successful generation.
-From these results, the abstract has been submitted for a conference and we are in a process of completing the first draft of journal paper.

Strategy for Future Research Activity

- We still have to collect more patient datasets.
- Perform k-fold cross validation and more analysis.
- Evaluate the performance of our networks using data obtained from different machine.
- Present and publish the result.

Causes of Carryover

We already had a GPU workstation computer available this year, so the money will be used in buying a better version according to the progress of the research and on paper publishing and conference attendance