Related Experiment Video
Updated: Jun 19, 2025
09:07
Multi-Tracer Studies of Brain Oxygen and Glucose Metabolism Using a Time-of-Flight Positron Emission Tomography-Computed Tomography Scanner
Published on: June 7, 2024
300
Development of an automated region-of-interest-setting method based on a deep neural network for brain perfusion
Taeko Tomimatsu1, Kosuke Yamashita1, Takumi Sakata1
1Graduate School of Health Sciences, Kumamoto University, Japan.
Asia Oceania Journal of Nuclear Medicine & Biology
|July 25, 2024
Summary
A deep convolutional neural network (DCNN) method was developed to automatically identify pulmonary artery (PA) and ascending aorta (AAo) regions for quantitative cerebral blood flow measurements. This DCNN-ROI method achieves accuracy comparable to manual methods, improving repeatability and reproducibility.
Area of Science:
- Medical Imaging
- Radiology
- Artificial Intelligence in Medicine
Background:
- Quantitative measurement of regional cerebral blood flow (rCBF) is crucial for diagnosing various neurological conditions.
- Existing methods like simple noninvasive microsphere (SIMS) and improved brain uptake ratio (IBUR) rely on accurate input function determination.
- Accurate input functions depend on precise identification of the pulmonary artery (PA) and ascending aorta (AAo) regions of interest (ROIs) from dynamic chest images.
Purpose of the Study:
- To develop and validate a novel deep convolutional neural network (DCNN) based method for automated setting of PA and AAo ROIs.
- To improve the accuracy, repeatability, and reproducibility of input function determination for rCBF quantification methods (SIMS and IBUR).
Main Methods:
- A U-Net architecture, a type of DCNN, was employed for segmenting PA and AAo candidate regions.
- The DCNN-ROI method was trained and tested on dynamic chest images from 290 patients (123I-IMP) and 108 patients (99mTc-ECD).
- Input functions were calculated by integrating the area under the curve (AUC) of time-activity curves from DCNN-identified ROIs and compared with manual ROI methods.
Main Results:
- The DCNN-ROI method achieved 100% coincidence in locating PA and AAo ROIs compared to manual methods.
- Strong correlations were observed between AUC counts derived from DCNN-ROI and manual ROI methods.
- The developed DCNN-ROI method demonstrated comparable accuracy to manual ROI setting for input function determination.
Conclusions:
- A novel DCNN-ROI method was successfully developed for automated PA and AAo ROI setting.
- This automated method enhances the accuracy and reliability of input function determination for SIMS and IBUR rCBF quantification techniques.
- The DCNN-ROI method offers a reproducible and accurate alternative to manual ROI setting in quantitative SPECT imaging.
