Learning a high-order connectional brain template (CBT) endowed with cognitive capacities such as visual or auditory memory is crucial for identifying cognition-related biomarkers and distinguishing between control and clinical populations. Higher-order CBTs provide a population-level representation that captures not only structural or topological regularities but also the multi-regional interactions and cognitive processes that conventional pairwise models fail to reflect. Because the brain operates through complex, coordinated dynamics, estimating CBTs that incorporate such higher-order and cognitively meaningful organization is essential for advancing our understanding of neural function and dysfunction. While recent machine-learning and graph-neural-network approaches have improved CBT estimation, they remain limited by their focus on pairwise interactions and purely structural features, overlooking both higher-order organization and cognitive properties. This gap raises a central question: How can we learn a high-order CBT that is well-centered at the population level and also endowed with cognitive capacities? We tackle this challenge using reservoir computing (RC), a biologically inspired framework that mimics how the brain processes information. RC exhibits dynamic properties similar to those of the prefrontal cortex, an area associated with working memory and features a fading memory mechanism, known as the Echo State Property (ESP), which mirrors the brain's short-term memory function. Building on these properties, we introduce HyperCOCO, a novel framework for generating high-order cognitively enhanced CBTs in two stages. First, BOLD signals are processed through a random reservoir to generate high-order individual functional connectomes, which are then aggregated into a population-level template. Second, this template is instantiated into a hyper-cognitive reservoir and stimulated with multi-sensory inputs (visual, auditory, and linguistic). Finally, we measure the memory capacity of the resulting CBT as a proxy for its ability to encode and retain cognitive information. Our source code is available at https://github.com/basiralab/HyperCOCO.

Message passing is a core mechanism in Graph Neural Networks (GNNs), enabling the iterative update of node embeddings by aggregating information from neighboring nodes. Graph Convolutional Networks (GCNs) exemplify this approach by adapting convolutional operations for graph structures, allowing features from adjacent nodes to be combined effectively. However, GCNs encounter challenges with complex or dynamic data. Capturing long-range dependencies often requires deeper layers, which not only increase computational costs but also lead to over-smoothing, where node embeddings become indistinguishable. To overcome these challenges, reservoir computing has been integrated into GNNs, leveraging iterative message-passing dynamics for stable information propagation without extensive parameter tuning. Despite its promise, existing reservoir-based models lack structured convolutional mechanisms, limiting their ability to accurately aggregate multi-hop neighborhood information. To address these limitations, we propose RGC-Net (\emph{Reservoir-based Graph Convolutional Network}), which integrates reservoir dynamics with structured graph convolution. Key contributions include: (i) a reimagined convolutional framework with fixed-random reservoir weights and a leaky integrator to enhance feature retention; (ii) a robust, adaptable model for graph classification; and (iii) an RGC-Net-powered transformer for graph generation with application to dynamic brain connectivity. Extensive experiments show RGC-Net achieves state-of-the-art performance in classification and generative tasks, including brain graph evolution, with faster convergence and mitigated over-smoothing. Our source code is available at https://github.com/basiralab/RGC-Net.

A connectional brain template (CBT) is a holistic representation of a population of brain connectivities. The federated learning of CBT allows for estimating the CBT of brain connectivities from multiple domains (i.e., hospitals) in a fully data-preserving manner. However, existing methods overlook the non-independent and identically distributed (non-IID) issue stemming from the heterogeneity of multi-domain brain connectivities. This non-IID issue degrades the centrality of locally learned CBT from multiple decentralized domains, eventually leading to a limited representation ability. To overcome this limitation, we propose a metadata-driven federated learning framework, called MetaFedCBT, for multi-domain CBT learning under the non-IID condition. Given the data drawn from a specific domain, our model is able to predict the metadata (i.e., statistics) of other unseen domains with a proposed metadata regressor and local-global network residual weights. Furthermore, we introduce a metadata-driven connectivity generator to predict brain connectivities of unseen domains under the guidance of obtained metadata. As the federated learning progresses over multiple rounds, we continuously update the predicted metadata and brain connectivities to better approximate the unseen domains. MetaFedCBT overcomes the non-IID issue by generating informative brain connectivities for privacy-preserving holistic CBT learning. Extensive experiments on multi-view morphological brain networks of normal and patient subjects demonstrate that our MetaFedCBT is a superior federated CBT learning model and significantly advances state-of-the-art performance.

The understanding of the convoluted evolution of infant brain networks during the first postnatal year is pivotal for identifying the dynamics of early brain connectivity development. Thanks to the valuable insights into the brain's anatomy, existing deep learning frameworks focused on forecasting the brain evolution trajectory from a single baseline observation. While yielding remarkable results, they suffer from three major limitations. First, they lack the ability to generalize to multi-trajectory prediction tasks, where each graph trajectory corresponds to a particular imaging modality or connectivity type (e.g., T1-w MRI). Second, existing models require extensive training datasets to achieve satisfactory performance which are often challenging to obtain. Third, they do not efficiently utilize incomplete time series data. To address these limitations, we introduce FedGmTE-Net++, a federated graph-based multi-trajectory evolution network. Using the power of federation, we aggregate local learnings among diverse hospitals with limited datasets. As a result, we enhance the performance of each hospital's local generative model, while preserving data privacy. The three key innovations of FedGmTE-Net++ are: (i) presenting the first federated learning framework specifically designed for brain multi-trajectory evolution prediction in a data-scarce environment, (ii) incorporating an auxiliary regularizer in the local objective function to exploit all the longitudinal brain connectivity within the evolution trajectory and maximize data utilization, (iii) introducing a two-step imputation process, comprising a preliminary K-Nearest Neighbours based precompletion followed by an imputation refinement step that employs regressors to improve similarity scores and refine imputations. Our comprehensive experimental results showed the outperformance of FedGmTE-Net++ in brain multi-trajectory prediction from a single baseline graph in comparison with benchmark methods. Our source code is available at https://github.com/basiralab/FedGmTE-Net-plus.

Learning from limited data has been extensively studied in machine learning, considering that deep neural networks achieve optimal performance when trained using a large amount of samples. Although various strategies have been proposed for centralized training, the topic of federated learning with small datasets remains largely unexplored. Moreover, in realistic scenarios, such as settings where medical institutions are involved, the number of participating clients is also constrained. In this work, we propose a novel federated learning framework, named RepTreeFL. At the core of the solution is the concept of a replica, where we replicate each participating client by copying its model architecture and perturbing its local data distribution. Our approach enables learning from limited data and a small number of clients by aggregating a larger number of models with diverse data distributions. Furthermore, we leverage the hierarchical structure of the clients network (both original and virtual), alongside the model diversity across replicas, and introduce a diversity-based tree aggregation, where replicas are combined in a tree-like manner and the aggregation weights are dynamically updated based on the model discrepancy. We evaluated our method on two tasks and two types of data, graph generation and image classification (binary and multi-class), with both homogeneous and heterogeneous model architectures. Experimental results demonstrate the effectiveness and outperformance of RepTreeFL in settings where both data and clients are limited.

Graph neural network (GNN) ushered in a new era of machine learning with interconnected datasets. While traditional neural networks can only be trained on independent samples, GNN allows for the inclusion of intersample interactions in the training process. This gain, however, incurs additional memory cost, rendering most GNNs unscalable for real-world applications involving vast and complicated networks with tens of millions of nodes (e.g., social circles, web graphs, and brain graphs). This means that storing the graph in the main memory can be difficult, let alone training the GNN model with significantly less GPU memory. While much of the recent literature has focused on either mini-batching GNN methods or quantization, graph reduction methods remain largely scarce. Furthermore, present graph reduction approaches have several drawbacks. First, most graph reduction focuses only on the inference stage (e.g., condensation, pruning, and distillation) and requires full graph GNN training, which does not reduce training memory footprint. Second, many methods focus solely on the graph's structural aspect, ignoring the initial population feature-label distribution, resulting in a skewed postreduction label distribution. Here, we propose a feature-label constrained graph net collapse (FALCON) to address these limitations. Our three core contributions lie in: 1) designing FALCON, a topology-aware graph reduction technique that preserves feature-label distribution by introducing a K-means clustering with a novel dimension-normalized Euclidean distance; 2) implementation of FALCON with other state-of-the-art (SOTA) memory reduction methods (i.e., mini-batched GNN and quantization) for further memory reduction; and 3) extensive benchmarking and ablation studies against SOTA methods to evaluate FALCON memory reduction. Our comprehensive results show that FALCON can significantly collapse various public datasets (e.g., PPI and Flickr to as low as 34% of the total nodes) while keeping equal prediction quality across GNN models. Our FALCON code is available at https://github.com/basiralab/FALCON.

In recent years, Connectional Brain Templates (CBTs) have become essential tools in network neuroscience for the representation of neural connections across populations. These graph-based representations are invaluable for comparing brain connectivity across individuals and identifying deviations associated with neurological and psychiatric disorders. Traditional methods for creating CBTs, such as linear averaging, struggle to capture the non-linear relationships within brain networks, limiting their effectiveness. The Deep Graph Normalizer (DGN), although it has been effective in fusing multi-view brain networks, but encounters significant challenges when the size of the data grows, thus leading to scalability issues and memory bottlenecks. These limitations restrict DGN’s applicability to large-scale brain networks, hindering further progress in the field. To address these challenges, we propose the Sparse-Clustered Deep Graph Normalizer Network (SCDGN), an enhanced version of DGN designed to improve scalability and computational efficiency. SCDGN integrates sparsification and hierarchical clustering within the DGN framework, enabling it to learn a cluster assignment matrix over nodes using the output of a GNN model. This approach allows SCDGN to process large graphs more effectively, reducing memory usage by 35% compared to DGN on normal and patient datasets while maintaining computational efficiency. Our experimental results demonstrate that SCDGN can handle large-scale connectomic datasets, offering a robust solution for estimating CBTs in complex brain networks.

The open neuroimaging datasets provided by researchers offer a wealth of samples for scientific research, enhancing reproducibility and accelerating new scientific discoveries. However, due to privacy concerns and the costs of data management, researchers often release data that has been processed using atlases. Nevertheless, releasing such data has some limitations, especially in the field of connectomics. Different studies may use different atlases, leading to brain connectivity data that is not directly comparable across studies. Additionally, since there is no universally accepted standard atlas, researchers have to compromise on atlas selection, which may not meet the needs of all studies. To address these limitations, we propose a cross-atlas brain connectivity mapping framework based on a dual-conditional diffusion model, which can generate brain connectivity corresponding to a target atlas given only the brain connectivity corresponding to an original atlas. We introduce the first deep learning framework for cross-atlas brain connectivity mapping and demonstrate its effectiveness through experiments. We also validate the effectiveness of the dual-conditional diffusion model through ablation experiments, showing that adding additional conditional information provides a richer source of guidance.

The rapid advancements in deep learning technologies have profoundly influenced the field of medical image analysis, yet their full integration into clinical radiology practices has not progressed as quickly as expected. A significant hurdle to their widespread adoption among radiologists and clinicians is the prevailing lack of trust and confidence in the outcomes produced by these technologies. This concern primarily stems from concerns regarding the explainability and generalizability of deep learning models within the realm of medical imaging. As part of the responses from the Medical Image Analysis Community to address these critical issues, we organized the IEEE Transactions on Neural Networks and Learning Systems (TNNLS) Special Issue on explainable and generalizable deep learning for medical imaging. This IEEE TNNLS Special Issue calls for original and innovative methodological contributions that aim to address the key challenges on explainability and generalizability of deep learning for medical imaging. This IEEE TNNLS Special Issue emphasizes the research and advanced development of the technical aspects of new image analysis methodologies, and all the developed new methods should also be evaluated or validated on real and large-scale medical imaging data.

Several machine learning approaches have been proposed to automatically derive clinical phenotypes from patient data. Nevertheless, methods leveraging similarity-based patient networks remain underexplored for temporal data. In this work, we propose a graph neural network (GNN) model that learns patient representation using different network configurations and feature modes. To explore the sequential nature of time series, features were extracted using a recurrent neural network (RNN) and embedded using information from the network structure via the GNN. Our method improves upon statistical and RNN baselines, with performance boosts up to 1% and 22% accuracy in the inductive and transductive settings, respectively. We also show that network configurations significantly impact performance in the transductive learning setting. Thus, automated phenotyping models based on GNNs could be used to support phenotype-based clinical research and ultimately for personalized clinical decision support.Data and Code Availability: This paper uses the MIMIC-III dataset [1], which is available on the PhysioNet repository [2]. The experiments are based on the public open source phenotyping benchmark of Harutyunyan et al. [3]. All our source code is publicly available at https://github.com/ds4dh/mimic3-benchmarks-GraDSCI23.

Noninvasive medical neuroimaging has yielded many discoveries about the brain connectivity. Several substantial techniques mapping morphological, structural and functional brain connectivities were developed to create a comprehensive road map of neuronal activities in the human brain -namely brain graph. Relying on its non-euclidean data type, graph neural network (GNN) provides a clever way of learning the deep graph structure and it is rapidly becoming the state-of-the-art leading to enhanced performance in various network neuroscience tasks. Here we review current GNN-based methods, highlighting the ways that they have been used in several applications related to brain graphs such as missing brain graph synthesis and disease classification. We conclude by charting a path toward a better application of GNN models in network neuroscience field for neurological disorder diagnosis and population graph integration. The list of papers cited in our work is available at https://github.com/basiralab/GNNs-in-Network-Neuroscience.

Foreseeing the evolution of brain connectivity between anatomical regions from a baseline observation can propel early disease diagnosis and clinical decision making. Such task becomes challenging when learning from multiple decentralized datasets with missing timepoints (e.g., datasets collected from different hospitals with a varying sequence of acquisitions). Federated learning (FL) is an emerging paradigm that enables collaborative learning among multiple clients (i.e., hospitals) in a fully privacy-preserving fashion. However, to the best of our knowledge, there is no FL work that foresees the time-dependent brain connectivity evolution from a single timepoint-let alone learning from non-iid decentralized longitudinal datasets with varying acquisition timepoints. In this paper, we propose the first FL framework to significantly boost the predictive performance of local hospitals with missing acquisition timepoints while benefiting from other hospitals with available data at those timepoints without sharing data. Specifically, we introduce 4D-FED-GNN+, a novel longitudinal federated GNN framework that works in (i) a uni-mode, where it acts as a graph self-encoder if the next timepoint is locally missing or (ii) in a dual-mode, where it concurrently acts as a graph generator and a self-encoder if the local follow-up data is available. Further, we propose a dual federation strategy, where (i) GNN layer-wise weight aggregation and (ii) pairwise GNN weight exchange between hospitals in a random order. To improve the performance of the poorly-conditioned hospitals (e.g., consecutive missing timepoints, intermediate missing timepoint), we further propose a second variant, namely 4D-FED-GNN++, which federates based on an ordering of the local hospitals computed using their incomplete sequential patterns. Our comprehensive experiments on real longitudinal datasets show that overall 4D-FED-GNN+ and 4D-FED-GNN++ significantly outperform benchmark methods. Our source code is available at https://github.com/basiralab/4D-FedGNN-Plus.

The field of network neuroscience provided unprecedented insights into how brain connectivity gets altered by autism spectrum disorder (ASD) on functional, structural, and morphological levels. However, a few studies have looked to design a framework that captures the complex network structure of the brain and disentangles the heterogeneity of ASD. In this paper, we leverage multi-kernel unsupervised learning in the construction of multiview hypergraph neural networks (HGNN), each capturing a particular view of the brain connectome, to eventually distinguish between ASD and normal control (NC) subjects. Additionally, we tested and measured how our proposed framework compares to other variants based on previous baseline methods. Our classification results outperformed comparison methods and agreed with the literature in the sense that the right hemisphere connectivity was more discriminative in ASD diagnosis than the left hemisphere.

Brain cancer phenotyping and treatment is highly informed by radiomic analyses of medical images. Specifically, the reliability of radiomics, which refers to extracting features from the tumor image intensity, shape and texture, depends on the accuracy of the tumor boundary segmentation. Hence, developing fully-automated brain tumor segmentation methods is highly desired for processing large imaging datasets. In this work, we propose a cooperative learning framework for multi-label brain tumor segmentation, which leverages on Structured Random Forest (SRF) and Bayesian Networks (BN). Basically, we embed both strong SRF and BN classifiers into a multi-layer deep architecture, where they cooperate to better learn tumor features for our multi-label classification task. The proposed SRF-BN cooperative learning integrates two complementary merits of both classifiers. While, SRF exploits structural and contextual image information to perform classification at the pixel-level, BN represents the statistical dependencies between image components at the superpixel-level. To further improve this SRF-BN cooperative learning, we ‘deepen’ this cooperation through proposing a multilayer framework, wherein each layer, BN inputs the original multi-modal MR images along with the probability maps generated by SRF. Through transfer learning from SRF to BN, the performance of BN improves. In turn, in the next layer, SRF will also benefit from the learning of BN through inputting the BN segmentation maps along with the original multimodal images. With the exception of the first layer, both classifiers use the output segmentation maps resulting from the previous layer, in the spirit of auto context models. We evaluated our framework on 50 subjects with multimodal MR images (FLAIR, T1, T1-c) to segment the whole tumor, its core and enhanced tumor. Our segmentation results outperformed those of several comparison methods, including the independent (non-cooperative) learning of SRF and BN.