Graph Convolutional Network-based Functional-structural Sub-region Framework for Head and Neck Cancer Prognosis with PET/CT Imaging
- Resource Type
- Conference
- Authors
- Zhou, Zidong; Peng, Junyi; Lin, Guoyu; Wu, Huiqin; Lv, Wenbing; Lu, Lijun
- Source
- 2022 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC) Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), 2022 IEEE. :1-3 Nov, 2022
- Subject
- Components, Circuits, Devices and Systems
Computing and Processing
Engineered Materials, Dielectrics and Plasmas
Nuclear Engineering
Photonics and Electrooptics
Signal Processing and Analysis
Head
Decision making
Neck
Convolutional neural networks
Prognostics and health management
Radiomics
Cancer
Sub-region
Graph convolutional neural network
Prognosis
Head and neck cancer
- Language
- ISSN
- 2577-0829
In order to improve the behavior of the prognostic model for patients with head and neck cancer (HNC), we constructed a graph convolutional network (GCN), characterizing the interaction of functional and structural information among different sub-regions of tumors in both PET and CT images. Sub-regions were obtained by using the Watershed-based method and were regarded as vertexes in the graph. The element of 1 in the adjacency matrix indicates corresponding two vertices are spatially contiguous, while the element of 0 indicates they are separated. The selected radiomics features were used to characterize each vertex. 642 HNC patients from 8 different centers were used. The best C-index of 0.767 was achieved by the combination of GCN and clinical parameters compared to the radiomics model (0.747) or GCN model (0.702). the proposed approach has great potential in aiding treatment decision-making.