Author: | Tang, Chung Ting |
Title: | Deep learning to reduce scan time and radiation dose in myocardial perfusion imaging SPECT |
Advisors: | Yoo, Jung Sun (HTI) Qin, Harry (SN) |
Degree: | DHSc |
Year: | 2022 |
Subject: | Heart -- Tomography Heart -- Radionuclide imaging Deep learning (Machine learning) Hong Kong Polytechnic University -- Dissertations |
Department: | Faculty of Health and Social Sciences |
Pages: | xvii, 196 pages : color illustrations |
Language: | English |
Abstract: | Background Coronary heart disease (CHD) is the leading cause of morbidity and mortality worldwide and SPECT MPI remains one of the most commonly used diagnostic imaging examinations to assess the function of the heart muscle. Deep Learning is an emerging technology for a wide range of different applications, including computer vision, it uses deep learning network composed of many layers to extract feature from the images. Despite vibrant technological evolution of SPECT in the past decades, issues of prolonged scan time and relatively high radiation dose remain to be resolved. Aim To propose and evaluate a deep learning-based approach to reduce scan time and radiation dose in MPI SPECT imaging without compromising diagnostic accuracy. Methods The new DL-based approach was developed to generate full time (FT)/full dose (FD) images from reduced time/low dose images and evaluated for accurate diagnosis as compared with clinical evaluation parameters. Six hundred pairs of stress and rest Tc-99m sestamibi MPI images were collected retrospectively from Queen Elizabeth Hospital in Hong Kong. Simulation of half and one third time/dose dataset were performed by discarding 50% and 66.6% of projections profiles from FT/FD ground truth dataset. A Generative Adversarial Networks with Dual-Domain U-Net Based Discriminators (DUGAN) was used to predict FT/FD SPECT MPI images and compared with other state-of-the-art DL networks. Similarity between DL synthesized images and standard FT/FD images was evaluated using various metrics, i.e., maximum absolute error (MAE), root mean square error (RMSE) and structural similarity index (SSIM). To judge clinical performance, current gold standard parameters, i.e., total perfusion deficit (TPD), perfusion defect area (TDA), and perfusion defect extent (TDE), summed rest score (SRS) and summed stress score (SSS) were assessed using clinical nuclear cardiology software Results The proposed method demonstrates effectiveness to reduce scan time and radiation dose in MPI SPECT imaging. The generated FT/FD MPI images from the half time/dose images using DUGAN showed superior overall imaging performance achieving the MAE of 0.0042, RMSE of 0.2175, and SSIM of 0.9767 (p ≤ 0.05) for the rest dataset and the MAE of 0.0027, RMSE of 0.1874 and SSIM of 0.9902 (p ≤ 0.05) for the stress dataset. The results from the clinical evaluation also proved that the synthesized FT/FD images from the half time/dose images could provide similar clinical performance with the ground truth standard FT/FD images both for the rest and stress datasets (Pearson correlation coefficient of 0.986 for TPD and 0.991 SRS between the DL predicted images from half time/dose images and the reference standard images with stress MPIs). Conclusions Deep learning could transform SPECT MPI into a more efficient and radiation minimized examination. The results presented in this study demonstrates the feasibility of deep learning technique to reduce scan time and radiation dose at least into half of the standard technique in MPI SPECT imaging without compromising clinical imaging performance. This could pave the way for clinical translation to circumvent inherent technical problems of SPECT with long acquisition time and relatively high radiation and unleash the potential of deep learning to revolutionize the clinical practice of SPECT imaging. |
Rights: | All rights reserved |
Access: | restricted access |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
6683.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 7.73 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12251