Anomaly Detection in Human Brain via Inductive Learning on Temporal Multiplex Networks

Ali Behrouz, Margo Seltzer
Proceedings of the 8th Machine Learning for Healthcare Conference, PMLR 219:50-75, 2023.

Abstract

The human brain is at the center of complex neurobiological systems, and understanding its structural and functional mechanisms remains an intriguing goal for neuroscience research. While magnetic resonance imaging (MRI) is one of the most widespread and important sources of neurological data, it poses daunting analysis challenges. Due to recent advances in graph theory and machine learning on graphs, representing the connections of the human brain as a network has become one of the most pervasive analytical paradigms. However, most existing graph machine learning-based methods suffer from a subset of four critical limitations: They are 1) designed for one type of data (e.g., fMRI or sMRI) and one individual subject, limiting their ability to use complementary information provided by different images. 2) designed in supervised or transductive settings, limiting their generalizability to unseen patterns. 3) designed for classifying brain networks, limiting their ability to reveal underlying patterns that might cause the symptoms of a disease or disorder. 4) frequently unable to scale to large numbers of samples. To address the first limitation, we suggest using multiplex networks–networks with different types of connections– to model the network of different data samples. We present ADMire, an inductive and unsupervised anomaly detection method for multiplex brain networks that can detect anomalous patterns in the brains of people living with a disease or disorder. It uses two different casual multiplex walks, inter-view and intra-view, to automatically extract and learn temporal network motifs. It then uses an anonymization strategy to hide node identity, keeping the model inductive. We then propose a novel negative sample generator strategy for multiplex networks that lets our model learn anomalous patterns in an unsupervised manner. Our experiments on Parkinson’s Disease, Attention Deficit Hyperactivity Disorder, and Autism Spectrum Disorder show the efficiency and effectiveness of our approach in detecting anomalous brain activity in people living with these diseases or disorders.

Cite this Paper

BibTeX
@InProceedings{pmlr-v219-behrouz23a, title = {Anomaly Detection in Human Brain via Inductive Learning on Temporal Multiplex Networks}, author = {Behrouz, Ali and Seltzer, Margo}, booktitle = {Proceedings of the 8th Machine Learning for Healthcare Conference}, pages = {50--75}, year = {2023}, editor = {Deshpande, Kaivalya and Fiterau, Madalina and Joshi, Shalmali and Lipton, Zachary and Ranganath, Rajesh and Urteaga, Iñigo and Yeung, Serene}, volume = {219}, series = {Proceedings of Machine Learning Research}, month = {11--12 Aug}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v219/behrouz23a/behrouz23a.pdf}, url = {https://proceedings.mlr.press/v219/behrouz23a.html}, abstract = {The human brain is at the center of complex neurobiological systems, and understanding its structural and functional mechanisms remains an intriguing goal for neuroscience research. While magnetic resonance imaging (MRI) is one of the most widespread and important sources of neurological data, it poses daunting analysis challenges. Due to recent advances in graph theory and machine learning on graphs, representing the connections of the human brain as a network has become one of the most pervasive analytical paradigms. However, most existing graph machine learning-based methods suffer from a subset of four critical limitations: They are 1) designed for one type of data (e.g., fMRI or sMRI) and one individual subject, limiting their ability to use complementary information provided by different images. 2) designed in supervised or transductive settings, limiting their generalizability to unseen patterns. 3) designed for classifying brain networks, limiting their ability to reveal underlying patterns that might cause the symptoms of a disease or disorder. 4) frequently unable to scale to large numbers of samples. To address the first limitation, we suggest using multiplex networks–networks with different types of connections– to model the network of different data samples. We present ADMire, an inductive and unsupervised anomaly detection method for multiplex brain networks that can detect anomalous patterns in the brains of people living with a disease or disorder. It uses two different casual multiplex walks, inter-view and intra-view, to automatically extract and learn temporal network motifs. It then uses an anonymization strategy to hide node identity, keeping the model inductive. We then propose a novel negative sample generator strategy for multiplex networks that lets our model learn anomalous patterns in an unsupervised manner. Our experiments on Parkinson’s Disease, Attention Deficit Hyperactivity Disorder, and Autism Spectrum Disorder show the efficiency and effectiveness of our approach in detecting anomalous brain activity in people living with these diseases or disorders.} }
Endnote
%0 Conference Paper %T Anomaly Detection in Human Brain via Inductive Learning on Temporal Multiplex Networks %A Ali Behrouz %A Margo Seltzer %B Proceedings of the 8th Machine Learning for Healthcare Conference %C Proceedings of Machine Learning Research %D 2023 %E Kaivalya Deshpande %E Madalina Fiterau %E Shalmali Joshi %E Zachary Lipton %E Rajesh Ranganath %E Iñigo Urteaga %E Serene Yeung %F pmlr-v219-behrouz23a %I PMLR %P 50--75 %U https://proceedings.mlr.press/v219/behrouz23a.html %V 219 %X The human brain is at the center of complex neurobiological systems, and understanding its structural and functional mechanisms remains an intriguing goal for neuroscience research. While magnetic resonance imaging (MRI) is one of the most widespread and important sources of neurological data, it poses daunting analysis challenges. Due to recent advances in graph theory and machine learning on graphs, representing the connections of the human brain as a network has become one of the most pervasive analytical paradigms. However, most existing graph machine learning-based methods suffer from a subset of four critical limitations: They are 1) designed for one type of data (e.g., fMRI or sMRI) and one individual subject, limiting their ability to use complementary information provided by different images. 2) designed in supervised or transductive settings, limiting their generalizability to unseen patterns. 3) designed for classifying brain networks, limiting their ability to reveal underlying patterns that might cause the symptoms of a disease or disorder. 4) frequently unable to scale to large numbers of samples. To address the first limitation, we suggest using multiplex networks–networks with different types of connections– to model the network of different data samples. We present ADMire, an inductive and unsupervised anomaly detection method for multiplex brain networks that can detect anomalous patterns in the brains of people living with a disease or disorder. It uses two different casual multiplex walks, inter-view and intra-view, to automatically extract and learn temporal network motifs. It then uses an anonymization strategy to hide node identity, keeping the model inductive. We then propose a novel negative sample generator strategy for multiplex networks that lets our model learn anomalous patterns in an unsupervised manner. Our experiments on Parkinson’s Disease, Attention Deficit Hyperactivity Disorder, and Autism Spectrum Disorder show the efficiency and effectiveness of our approach in detecting anomalous brain activity in people living with these diseases or disorders.
APA
Behrouz, A. & Seltzer, M.. (2023). Anomaly Detection in Human Brain via Inductive Learning on Temporal Multiplex Networks. Proceedings of the 8th Machine Learning for Healthcare Conference, in Proceedings of Machine Learning Research 219:50-75 Available from https://proceedings.mlr.press/v219/behrouz23a.html.

Related Material