Automatic Archiving and Classification of Positron Emission Tomography Images Using Deep Learning Models at Different Scan Times Publisher



Ghafari A¹ ; Sheikhzadeh P² Show Affiliations
Source: 2021 IEEE Nuclear Science Symposium and Medical Imaging Conference Record# NSS/MIC 2021 and 28th International Symposium on Room-Temperature Semiconductor Detectors# RTSD 2022 Published:2021

Abstract

Positron Emission Tomography provides physicians with important metabolic and semi-quantitative information. With the growing interest in applications of deep learning methods in medical imaging, a tool for archiving and categorizing images based on the body region is needed for the data preparation phase of the deep learning applications. We developed deep learning models for classifying images based on the body region (Quadratic Classification) and radiopharmaceuticals used in imaging (Binary Classification) for different scan times. Sensitivity and Specificity for Binary Classification (68Ga-PSMA and 18F-FDG) are in the range of 0.9933 - 0.9990. For the Quadratic Classification of 18F-FDG, Sensitivity and Specificity are 0.9873 ± 0.0114 and 0.9958± 0.0038, respectively. Likewise, the sensitivity and specificity of 68Ga-PSMA are 0.9788± 0.0123 and 0.9929 ± 0.0041, respectively. Results are promising and are indicative of the models' capability as a complementary tool for archiving and classifying images for further investigations. © 2021 IEEE.