About

The rapid growth of the aging population and the increasing prevalence of age-related neurodegenerative disorders, particularly Alzheimer's disease (AD), Parkinsons; and developmental disorders such as Autism, and attention deficit hyperactivity disorder (ADHD) pose significant challenges to public health and society. The recent advancements in neuroimaging data acquisition and several open-source data acquisition initiatives, primarily originating from North America and Europe, have enabled us to understand these disorders. Recent efforts in South Asia are directed toward acquisition of such large-scale data sets to understand the differences in these demographics. These neuroimaging datasets, including structural/functional magnetic resonance imaging (sMRI, fMRI), diffusion tensor imaging (DTI), electroencephalography (EEG), and magnetoencephalography (MEG), provide rich spatio-temporal details in brain structure and functioning. While sMRI and DTI give spatial and structural topography of the brain and the fiber tracts that connect across brain regions, fMRI, EEG/ MEG provide dynamic functional brain activity corresponding to the brain function. However, the data analytic methods to analyze these complex multimodal datasets, in the South Asian context, have not been adequately developed. 

This workshop is aimed at bringing together researchers across interdisciplinary fields to work collectively on advancements in analytical methods and data analytics to address challenges in dealing with multimodal neuroimaging datasets. The objective is to introduce the audience to the availability of multimodal neuroimaging data, and the challenges they pose and to inspire them to develop new techniques to arrive at robust inferences on brain function/dysfunction. The datasets include multimodal datasets from the Human Connectome Project, UK Biobank, Adolescent Brain Cognitive Development, etc. The availability of opensource datasets gives rise to numerous opportunities for developing newer methods beyond the traditional univariate, multivariate data-driven techniques. However, there are several shortcomings: the differences in data acquisition protocols are not uniform and acquisition settings are also varied, for instance, the scanner settings and sampling rates are not the same across the datasets. Such variabilities hamper the seamless pooling of data. Therefore, there is a need to harmonize the datasets with better signal processing techniques to address these shortcomings. These advances would enable us to effectively use these datasets for better training of state-of-the-art explainable deep learning/neural network models and for generalizability. The challenges faced in neuroimaging data analyses are domain-specific and different from those seen in traditional areas such as speech, radar, image processing, or computer vision. Multimodal neuroimaging datasets provide complementary information about brain function/dysfunction. However, they are complex with different spatial and temporal resolutions, which warrant the development of advanced analytical methods. One of the aims of the workshop is to familiarize the participants from a traditional signal/speech/image processing background to the unique challenges posed by neuroimaging data analysis.

In keeping with the main conference theme of celebration of signal processing, this workshop intends to bring in a blend of traditional signal processing techniques with the advancements in machine learning and artificial intelligence, to develop more robust, interdisciplinary analysis techniques for handling multimodal neuroimaging data. Its relevance lies in i) introducing neuroimaging data and its unique challenges and ii) sharpening the traditional tools with recent advancements, insights, and broader potential of adaptability and interpretability for neuroimaging applications. The workshop is intended to pool together researchers with varied domains of expertise, such as signal processing, machine learning, topological analysis, and statistics, all of which are critical in neuroimage analysis. Presentations of novel ideas by selected researchers, exposure to the state-of-the-art by keynote speakers, delineation of challenges, and a roadmap for the future in a panel discussion will set the tone of the research ahead in this area. 

Advancements in various domains are actively being applied in neuroimaging, yielding superior results. Challenges in tasks such as classification, segmentation, registration, and super-resolution in structural MR images are now being addressed using advanced machine learning techniques such as few-shot learning, selfsupervised learning, and latent embeddings. The analysis of fMRI time series with traditional univariate statistical analyses has moved on to modern deep learning methods such as convolutional neural networks, recurrent neural networks, and foundation models. Also, concepts of non-linear analysis such as recurrence plots, persistent homology from topological analysis, and graph-theoretic ideas are proving to be useful in understanding the disintegration of structures and functional networks, observed in neurodegeneration. Besides being robust, these approaches are also interpretable and require smaller data and lower computing resources for model-building. These developments motivate the incorporation of interdisciplinary approaches for analyzing neuroimaging data. Thus, this is the most suitable forum for enhancing outreach to allied areas in signal/image processing with an emphasis on solving important problems in neuroimaging. This workshop will also strengthen the interaction between the IEEE Signal Processing Society and newer IEEE-wide endeavors such as the IEEE Brain initiative, which will keep the momentum of the workshop going. Challenges based on multimodal data from less-studied regions, despite being populous, such as Indian and South Asian countries require continued future efforts.