A convolutional neural network-based system to estimate the arterial plasma radioactivity curve in 18 F-FDG dynamic brain PET study

Keisuke Kawauchi ¹, Mui Saito ¹, Kentaro Nishigami ¹

⁰Graduate School of Biomedical Science and Engineering, Hokkaido University.

Nuclear Medicine Communications +

|

August 29, 2023

View abstract on PubMed

Summary

This study introduces a convolutional neural network (CNN) system that noninvasively estimates arterial plasma input functions and quantifies cerebral metabolic rate of glucose (CMRGlc) from dynamic PET scans, eliminating the need for blood sampling.

Area Of Science

Medical Imaging
Nuclear Medicine
Computational Neuroscience

Background

Dynamic Positron Emission Tomography (PET) enables quantitative imaging of metabolic activity.
Conventional methods require invasive arterial blood sampling for input function estimation, limiting noninvasive analysis.
Accurate input function is crucial for quantifying metabolic rates like cerebral metabolic rate of glucose (CMRGlc).

Purpose Of The Study

To develop a convolutional neural network (CNN) system for noninvasive estimation of arterial plasma time-radioactivity curves.
To accurately quantify CMRGlc directly from dynamic PET data using the developed CNN.
To eliminate the need for invasive blood sampling in dynamic PET studies.

Main Methods

Retrospective analysis of 29 patients with neurological disorders undergoing 18F-FDG-PET/CT.
Development of a CNN architecture to estimate arterial plasma time-radioactivity curves.
Validation of CNN-estimated input functions against arterial blood samples.

Main Results

The CNN accurately estimated the time-radioactivity curve with prediction errors within 10% in at least one frame for all patients.
A highly significant correlation coefficient of 0.99 was observed between CMRGlc values derived from sampled blood and CNN estimation.
The developed CNN system demonstrated high accuracy in quantifying CMRGlc.

Conclusions

A CNN can accurately determine arterial plasma input functions and CMRGlc from dynamic 18F-FDG PET data.
CNN facilitates noninvasive measurements, significantly advancing quantitative dynamic PET analysis.
This approach holds potential for broader applications in quantitative medical imaging analysis.