TY - CHAP
T1 - Multimodal Brain Age Estimation Using Interpretable Adaptive Population-Graph Learning
AU - Bintsi, Kyriaki Margarita
AU - Baltatzis, Vasileios
AU - Potamias, Rolandos Alexandros
AU - Hammers, Alexander
AU - Rueckert, Daniel
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
PY - 2023
Y1 - 2023
N2 - Brain age estimation is clinically important as it can provide valuable information in the context of neurodegenerative diseases such as Alzheimer’s. Population graphs, which include multimodal imaging information of the subjects along with the relationships among the population, have been used in literature along with Graph Convolutional Networks (GCNs) and have proved beneficial for a variety of medical imaging tasks. A population graph is usually static and constructed manually using non-imaging information. However, graph construction is not a trivial task and might significantly affect the performance of the GCN, which is inherently very sensitive to the graph structure. In this work, we propose a framework that learns a population graph structure optimized for the downstream task. An attention mechanism assigns weights to a set of imaging and non-imaging features (phenotypes), which are then used for edge extraction. The resulting graph is used to train the GCN. The entire pipeline can be trained end-to-end. Additionally, by visualizing the attention weights that were the most important for the graph construction, we increase the interpretability of the graph. We use the UK Biobank, which provides a large variety of neuroimaging and non-imaging phenotypes, to evaluate our method on brain age regression and classification. The proposed method outperforms competing static graph approaches and other state-of-the-art adaptive methods. We further show that the assigned attention scores indicate that there are both imaging and non-imaging phenotypes that are informative for brain age estimation and are in agreement with the relevant literature.
AB - Brain age estimation is clinically important as it can provide valuable information in the context of neurodegenerative diseases such as Alzheimer’s. Population graphs, which include multimodal imaging information of the subjects along with the relationships among the population, have been used in literature along with Graph Convolutional Networks (GCNs) and have proved beneficial for a variety of medical imaging tasks. A population graph is usually static and constructed manually using non-imaging information. However, graph construction is not a trivial task and might significantly affect the performance of the GCN, which is inherently very sensitive to the graph structure. In this work, we propose a framework that learns a population graph structure optimized for the downstream task. An attention mechanism assigns weights to a set of imaging and non-imaging features (phenotypes), which are then used for edge extraction. The resulting graph is used to train the GCN. The entire pipeline can be trained end-to-end. Additionally, by visualizing the attention weights that were the most important for the graph construction, we increase the interpretability of the graph. We use the UK Biobank, which provides a large variety of neuroimaging and non-imaging phenotypes, to evaluate our method on brain age regression and classification. The proposed method outperforms competing static graph approaches and other state-of-the-art adaptive methods. We further show that the assigned attention scores indicate that there are both imaging and non-imaging phenotypes that are informative for brain age estimation and are in agreement with the relevant literature.
KW - Adaptive graph learning
KW - Brain age regression
KW - Graph Convolutional Networks
KW - Interpretability
UR - http://www.scopus.com/inward/record.url?scp=85174700456&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-43993-3_19
DO - 10.1007/978-3-031-43993-3_19
M3 - Conference paper
AN - SCOPUS:85174700456
SN - 9783031439926
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 195
EP - 204
BT - Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 - 26th International Conference, Proceedings
A2 - Greenspan, Hayit
A2 - Greenspan, Hayit
A2 - Madabhushi, Anant
A2 - Mousavi, Parvin
A2 - Salcudean, Septimiu
A2 - Duncan, James
A2 - Syeda-Mahmood, Tanveer
A2 - Taylor, Russell
PB - Springer Science and Business Media Deutschland GmbH
T2 - 26th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2023
Y2 - 8 October 2023 through 12 October 2023
ER -