(Sunday, 7/14/2019)
Room: ePoster Forums
Purpose: There are several main methods for image reconstruction technique for nuclear medicine. (ex. FBP, MLEM, OSEM, etc.) In order to overcome disadvantages of conventional technique such as noise, weaken signal, we proposed the image acquisition technique using deep profile learning.
Methods: Nuclear medicine image of 120 cases have been acquired by using Monte Carlo simulation. Over 10,000 profiles have been measured from all images to train the image acquisition system. And we prepared the profiles of ideal pattern from ground truth image regarding each reconstructed image. After the training with the profiles (X-label) from reconstructed image and ideal profile (Y-label) through the deep neural network using SoftMax. we inserted SINOGRAM of new cases to image acquisition system to get image.
Results: The deep neural network exported optimized profile from new SINOGRAM. We re-arranged these optimized profiles according to original order. The re-ordered profiles matrix can show the optimized image which has better image quality than the performance of conventional image reconstruction technique.
Conclusion: This study shows a feasibility about the image acquisition using the deep profile learning for nuclear medicine imaging with good performance.
Keywords
Monte Carlo, Reconstruction, Nuclear Medicine
Taxonomy
IM- PET : Machine learning, computer vision
Contact Email
