Encrypted login | home

Program Information

Application of Deep Learning to Nuclear Medicine Image Acquisition

no image available
H Shin

H Shin1*, D Yoon2 , T Suh3 , M Kim4 , J Jung5 , S Kim6 , H Yang7 , (1) Catholic University of Korea, Seoul, Seoul, (2) Catholic University of Korea, Seoul, Seoul, (3) Catholic Univ Medical College, Seoul, SEOUL, (4) Catholic University of Korea, Seoul, Seoul, (5) University of Florida, Gainesville, FL, (6) Catholic University of Korea, Seoul, Seoul, (7) ,


SU-K-601-10 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: 601

Purpose: The conventional iteration algorithm based on stochastic such as ordered subset expectation maximization (OSEM) generally lost the information of small region. Therefore, this signal recovery in small region is essential to diagnose the lesion using nuclear medicine imaging. The proposed study is to apply deep learning algorithm for fast image reconstruction using only profile information based on sinogram.

Methods: In order to compare the performance of our proposed method with conventional reconstruction method, we used the FBP and OSEM for nuclear medicine image reconstruction. The proposed algorithm was composed of four part to reconstruct nuclear medicine image using deep learning algorithm: Input, Supervised Learning and Convolutional Neural network (CNN), Weight decay, Output. The second part is used for profile learning and store the database for profile information. Moreover, this part is to learn the information of signal and noise region and optimize the profile. Next part is to recovery the signal of small regions using stochastic gradient descent method. It makes to increase true signal and decrease noise.

Results: We acquired the reconstructed image of FBP, OSEM and deep learning algorithm. The intensity of the true area including radiopharmaceutical was strengthened and the intensity noise area surrounding the true area was reduced. Therefore, the signal to noise ratio (SNR) of image using deep learning algorithm was enhanced by the signal of FBP and OSEM. The intensity from small region of OSEM was lost compared to that of the deep learning algorithm.

Conclusion: We confirmed that application of image reconstruction method using deep learning algorithm was successfully deducted. In conclusion, since this study could give the accurate and fast information and detect relatively small size tumor, it can realize an early diagnosis using nuclear medicine image.

Contact Email: