K-CapsNet: K-Nearest Neighbor Based Convolution Capsule Network for Cerenkov Luminescence Tomography Reconstruction.
Annu Int Conf IEEE Eng Med Biol Soc
; 2023: 1-4, 2023 07.
Article
en En
| MEDLINE
| ID: mdl-38082846
ABSTRACT
Cerenkov luminescence tomography (CLT) has received significant attention as a promising imaging modality that can display the three-dimensional (3D) distribution of radioactive probes. However, the reconstruction of CLT suffers from severe ill-posed problem. It is difficult for traditional model-based method to obtain satisfactory result. Recently, deep learning-based method have shown great potential for accurate and efficient CLT reconstruction. In this study, a KNN-based convolution capsule network, named K-CapsNet, is proposed for cerenkov luminescence tomography. In K-CapsNet, the surface photon intensity is encoded in capsule form. The KNN-based convolution and K-means clustering are proposed for efficient encoding. Numerical simulation experiments have been carried out to verify the performance of K-CapsNet, and the results show that it performs superior in source localization and morphological restoration compared with existing methods.
Texto completo:
1
Colección:
01-internacional
Base de datos:
MEDLINE
Asunto principal:
Tomografía Óptica
Idioma:
En
Revista:
Annu Int Conf IEEE Eng Med Biol Soc
Año:
2023
Tipo del documento:
Article