Pub Date : 2025-11-24DOI: 10.1109/MSP.2025.3590806
Malte Hoffmann
Deep learning has revolutionized neuroimage analysis by delivering unprecedented speed and accuracy. However, the narrow scope of many training datasets constrains model robustness and generalizability. This challenge is particularly acute in magnetic resonance imaging (MRI), where image appearance varies widely across pulse sequences and scanner hardware. A recent domain-randomization strategy addresses the generalization problem by training deep neural networks on synthetic images with randomized intensities and anatomical content. By generating diverse data from anatomical segmentation maps, the approach enables models to accurately process image types unseen during training, without retraining or fine-tuning. It has demonstrated effectiveness across modalities including MRI, computed tomography, positron emission tomography, and optical coherence tomography, as well as beyond neuroimaging in ultrasound, electron and fluorescence microscopy, and X-ray microtomography. This tutorial paper reviews the principles, implementation, and potential of the synthesis-driven training paradigm. It highlights key benefits, such as improved generalization and resistance to overfitting, while discussing trade-offs such as increased computational demands. Finally, the article explores practical considerations for adopting the technique, aiming to accelerate the development of generalizable tools that make deep learning more accessible to domain experts without extensive computational resources or machine learning knowledge.
{"title":"Domain-Randomized Deep Learning for Neuroimage Analysis: Selecting Training Strategies, Navigating Challenges, and Maximizing Benefits","authors":"Malte Hoffmann","doi":"10.1109/MSP.2025.3590806","DOIUrl":"10.1109/MSP.2025.3590806","url":null,"abstract":"Deep learning has revolutionized neuroimage analysis by delivering unprecedented speed and accuracy. However, the narrow scope of many training datasets constrains model robustness and generalizability. This challenge is particularly acute in magnetic resonance imaging (MRI), where image appearance varies widely across pulse sequences and scanner hardware. A recent domain-randomization strategy addresses the generalization problem by training deep neural networks on synthetic images with randomized intensities and anatomical content. By generating diverse data from anatomical segmentation maps, the approach enables models to accurately process image types unseen during training, without retraining or fine-tuning. It has demonstrated effectiveness across modalities including MRI, computed tomography, positron emission tomography, and optical coherence tomography, as well as beyond neuroimaging in ultrasound, electron and fluorescence microscopy, and X-ray microtomography. This tutorial paper reviews the principles, implementation, and potential of the synthesis-driven training paradigm. It highlights key benefits, such as improved generalization and resistance to overfitting, while discussing trade-offs such as increased computational demands. Finally, the article explores practical considerations for adopting the technique, aiming to accelerate the development of generalizable tools that make deep learning more accessible to domain experts without extensive computational resources or machine learning knowledge.","PeriodicalId":13246,"journal":{"name":"IEEE Signal Processing Magazine","volume":"42 4","pages":"78-90"},"PeriodicalIF":9.6,"publicationDate":"2025-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145593217","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-21DOI: 10.1109/MSP.2025.3605746
Morten Mørup;Evrim Acar;Tülay Adalı
Functional neuroimaging has become a central window into our working mind and how it changes by aging and disease. However, the different measurement modalities in functional neuroimaging are challenged by high dimensionality when compared with sample sizes. Furthermore, they exhibit high degrees of variability across individuals when analyzing multiset functional neuroimaging datasets. Whereas the multiple functional neuroimaging modalities currently available provide complementary views of brain function, their joint analysis remains an important challenge in neuroscience. This survey article highlights prominent modeling methodologies for the discovery of interpretable patterns in such multiset and multimodal functional neuroimaging datasets. The survey highlights prominent modeling strategies from hard to soft coupling for the modeling of these high-dimensional multiset and multimodal functional neuroimaging datasets while emphasizing the importance of model uniqueness as a prerequisite for reliable and reproducible pattern discovery. We also provide future directions of research for interpretable pattern discovery in functional neuroimaging that ultimately can further our understanding of perhaps one of nature’s most intriguing organs, the human brain.
{"title":"Tensor and Coupled Decompositions: Interpretable pattern discovery in multiset and multimodal functional neuroimaging data","authors":"Morten Mørup;Evrim Acar;Tülay Adalı","doi":"10.1109/MSP.2025.3605746","DOIUrl":"https://doi.org/10.1109/MSP.2025.3605746","url":null,"abstract":"Functional neuroimaging has become a central window into our working mind and how it changes by aging and disease. However, the different measurement modalities in functional neuroimaging are challenged by high dimensionality when compared with sample sizes. Furthermore, they exhibit high degrees of variability across individuals when analyzing multiset functional neuroimaging datasets. Whereas the multiple functional neuroimaging modalities currently available provide complementary views of brain function, their joint analysis remains an important challenge in neuroscience. This survey article highlights prominent modeling methodologies for the discovery of interpretable patterns in such multiset and multimodal functional neuroimaging datasets. The survey highlights prominent modeling strategies from hard to soft coupling for the modeling of these high-dimensional multiset and multimodal functional neuroimaging datasets while emphasizing the importance of model uniqueness as a prerequisite for reliable and reproducible pattern discovery. We also provide future directions of research for interpretable pattern discovery in functional neuroimaging that ultimately can further our understanding of perhaps one of nature’s most intriguing organs, the human brain.","PeriodicalId":13246,"journal":{"name":"IEEE Signal Processing Magazine","volume":"42 4","pages":"41-57"},"PeriodicalIF":9.6,"publicationDate":"2025-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145560711","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-21DOI: 10.1109/MSP.2025.3596731
Saurabh Sihag;Gonzalo Mateos;Alejandro Ribeiro
Neurodegenerative disorders exhibit correlated patterns of cortical atrophy across the brain, where the degree of atrophy deviates from what is expected in a typically healthy individual. Brain age is a data-driven estimate of biological age derived from neuroimaging datasets. An increasing brain age gap—defined as a higher predicted brain age relative to chronological age—can indicate greater vulnerability to neurodegeneration and cognitive decline. As such, the brain age gap is a promising biomarker for monitoring brain health. However, the practical adoption of brain age gap metrics derived from prevalent machine learning (ML) approaches is limited by various methodological obscurities that stem from opaque decision-making processes and insufficient handling of statistical phenomena inherent to neurodegeneration. This article introduces key mathematical principles for brain age gap prediction from the perspective of graph signal processing (GSP), aiming to address the core challenges hindering the practical use of the brain age gap as a biomarker. In this context, we survey a principled deep learning framework based on coVariance Neural Networks (VNNs) to infer the brain age gap from anatomical features extracted from structural neuroimaging. VNNs operate on the covariance matrix as a graph, and their theoretical foundations are inspired by recent advances in GSP. We demonstrate that brain age gap derived from a VNN-based ML pipeline exhibit stability, transferability across multi-scale datasets, and improved interpretability; all key properties that enhance the reproducibility and transparency required for principled brain age gap prediction in clinical applications.
{"title":"Disentangling Neurodegeneration With Brain Age Gap Prediction Models: A graph signal processing perspective","authors":"Saurabh Sihag;Gonzalo Mateos;Alejandro Ribeiro","doi":"10.1109/MSP.2025.3596731","DOIUrl":"https://doi.org/10.1109/MSP.2025.3596731","url":null,"abstract":"Neurodegenerative disorders exhibit correlated patterns of cortical atrophy across the brain, where the degree of atrophy deviates from what is expected in a typically healthy individual. Brain age is a data-driven estimate of biological age derived from neuroimaging datasets. An increasing brain age gap—defined as a higher predicted brain age relative to chronological age—can indicate greater vulnerability to neurodegeneration and cognitive decline. As such, the brain age gap is a promising biomarker for monitoring brain health. However, the practical adoption of brain age gap metrics derived from prevalent machine learning (ML) approaches is limited by various methodological obscurities that stem from opaque decision-making processes and insufficient handling of statistical phenomena inherent to neurodegeneration. This article introduces key mathematical principles for brain age gap prediction from the perspective of graph signal processing (GSP), aiming to address the core challenges hindering the practical use of the brain age gap as a biomarker. In this context, we survey a principled deep learning framework based on coVariance Neural Networks (VNNs) to infer the brain age gap from anatomical features extracted from structural neuroimaging. VNNs operate on the covariance matrix as a graph, and their theoretical foundations are inspired by recent advances in GSP. We demonstrate that brain age gap derived from a VNN-based ML pipeline exhibit stability, transferability across multi-scale datasets, and improved interpretability; all key properties that enhance the reproducibility and transparency required for principled brain age gap prediction in clinical applications.","PeriodicalId":13246,"journal":{"name":"IEEE Signal Processing Magazine","volume":"42 4","pages":"58-77"},"PeriodicalIF":9.6,"publicationDate":"2025-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145560715","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-21DOI: 10.1109/MSP.2025.3600011
Borbála Hunyadi;Selin Aviyente
Over the past decade, two complementary approaches have been proposed to detect the community structure of high-dimensional functional connectivity networks resulting from multiple modalities, time points, frequency bands or subjects. The first approach emerged from the field of network science where multilayer community detection algorithms such as maximizing multilayer modularity or minimizing the normalized cut have been proposed. The second approach emerged from the field of signal processing where tensors have been used to model high-dimensional networks where different low-rank tensor decomposition models are employed to reveal the underlying latent factors. While both research thrusts have provided valuable insight to the topology of brain networks, the equivalencies between the two approaches have not been studied in a systematic fashion up to date. This paper reviews the major community detection approaches for unraveling the topology of multilayer functional connectivity networks from the perspective of both network science and tensor decomposition. We show mathematical equivalencies between different tensor generative models and well-known graph partitioning objective functions. Finding these equivalencies can result in computationally efficient algorithms with optimality guarantees and inform the choice of different design parameters such as the number of communities.
{"title":"Brain Connectivity: From network science to tensor models","authors":"Borbála Hunyadi;Selin Aviyente","doi":"10.1109/MSP.2025.3600011","DOIUrl":"https://doi.org/10.1109/MSP.2025.3600011","url":null,"abstract":"Over the past decade, two complementary approaches have been proposed to detect the community structure of high-dimensional functional connectivity networks resulting from multiple modalities, time points, frequency bands or subjects. The first approach emerged from the field of network science where multilayer community detection algorithms such as maximizing multilayer modularity or minimizing the normalized cut have been proposed. The second approach emerged from the field of signal processing where tensors have been used to model high-dimensional networks where different low-rank tensor decomposition models are employed to reveal the underlying latent factors. While both research thrusts have provided valuable insight to the topology of brain networks, the equivalencies between the two approaches have not been studied in a systematic fashion up to date. This paper reviews the major community detection approaches for unraveling the topology of multilayer functional connectivity networks from the perspective of both network science and tensor decomposition. We show mathematical equivalencies between different tensor generative models and well-known graph partitioning objective functions. Finding these equivalencies can result in computationally efficient algorithms with optimality guarantees and inform the choice of different design parameters such as the number of communities.","PeriodicalId":13246,"journal":{"name":"IEEE Signal Processing Magazine","volume":"42 4","pages":"8-24"},"PeriodicalIF":9.6,"publicationDate":"2025-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145560708","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-21DOI: 10.1109/MSP.2025.3613224
Vince D. Calhoun;Damien Coyle;Javier Escudero;Borbala Hunyadi;Jing Sui
{"title":"Special Issue on Accelerating Brain Discovery Through Data Science and Neurotechnology: Part 1 [From the Guest Editors]","authors":"Vince D. Calhoun;Damien Coyle;Javier Escudero;Borbala Hunyadi;Jing Sui","doi":"10.1109/MSP.2025.3613224","DOIUrl":"https://doi.org/10.1109/MSP.2025.3613224","url":null,"abstract":"","PeriodicalId":13246,"journal":{"name":"IEEE Signal Processing Magazine","volume":"42 4","pages":"5-7"},"PeriodicalIF":9.6,"publicationDate":"2025-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11264270","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145560695","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-21DOI: 10.1109/MSP.2025.3613864
Yuntao Han;Themis Prodromakis;Shiwei Wang
Spike sorting facilitates neuroscientific research by isolating single-unit activity from extracellular recordings, enabling the study of individual neuron behavior. On-chip spike sorting systems address the data deluge challenge in high-channel-count neural probes by performing near-sensor processing directly on implanted hardware. Through hardware miniaturization, this approach reduces the need for bulky cabling and facilitates wireless data transmission, enabling chronic recordings in freely moving subjects. As neural probes scale to hundreds or thousands of channels, on-chip implementations pave the way for large-scale real-time analysis of neuronal ensembles and thus facilitate translating the technology from lab to more real-world applications.
{"title":"On-Chip Spike Sorting: Developments, challenges, and future directions","authors":"Yuntao Han;Themis Prodromakis;Shiwei Wang","doi":"10.1109/MSP.2025.3613864","DOIUrl":"https://doi.org/10.1109/MSP.2025.3613864","url":null,"abstract":"Spike sorting facilitates neuroscientific research by isolating single-unit activity from extracellular recordings, enabling the study of individual neuron behavior. On-chip spike sorting systems address the data deluge challenge in high-channel-count neural probes by performing near-sensor processing directly on implanted hardware. Through hardware miniaturization, this approach reduces the need for bulky cabling and facilitates wireless data transmission, enabling chronic recordings in freely moving subjects. As neural probes scale to hundreds or thousands of channels, on-chip implementations pave the way for large-scale real-time analysis of neuronal ensembles and thus facilitate translating the technology from lab to more real-world applications.","PeriodicalId":13246,"journal":{"name":"IEEE Signal Processing Magazine","volume":"42 4","pages":"103-117"},"PeriodicalIF":9.6,"publicationDate":"2025-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145560713","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：