Abstract:We propose a unified optimization framework that combines neural networks with dictionary learning to model complex interactions between resting state functional MRI and behavioral data. The dictionary learning objective decomposes patient correlation matrices into a collection of shared basis networks and subject-specific loadings. These subject-specific features are simultaneously input into a neural network that predicts multidimensional clinical information. Our novel optimization framework combines the gradient information from the neural network with that of a conventional matrix factorization objective. This procedure collectively estimates the basis networks, subject loadings, and neural network weights most informative of clinical severity. We evaluate our combined model on a multi-score prediction task using 52 patients diagnosed with Autism Spectrum Disorder (ASD). Our integrated framework outperforms state-of-the-art methods in a ten-fold cross validated setting to predict three different measures of clinical severity.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to model and predict multi - dimensional clinical features from functional connectomics data by integrating neural network and dictionary learning methods. Specifically, the researchers focus on how to use resting - state functional magnetic resonance imaging (rs - fMRI) data to predict multiple clinical severity indicators in patients with autism spectrum disorder (ASD). ### Background and Challenges - **Patient Heterogeneity**: Neurodevelopmental disorders such as autism and schizophrenia have great patient heterogeneity in clinical manifestations. - **High - Dimensional Data**: rs - fMRI data is high - dimensional and affected by external noise, which complicates the unified representation of behavior and connectivity data. - **Limitations of Existing Methods**: Most existing techniques mainly focus on distinguishing patients from control groups or locating functional regions in the brain, and few works can handle the problem of predicting continuous - value severity from connectomics data. ### Research Objectives - **Propose a New Optimization Framework**: Combine neural network and dictionary learning to establish a unified optimization framework to model the interaction between complex brain functional connectivity and behavioral data. - **Multi - Dimensional Clinical Prediction**: Use this framework to predict multiple clinical severity indicators and improve the accuracy and generalization ability of prediction. - **Interpretability and Flexibility**: Provide interpretability of the brain basis through dictionary learning, and at the same time use the learning ability of neural network to capture complex nonlinear relationships. ### Method Overview - **Dictionary Learning**: Decompose the patient's association matrix into a shared underlying network and an individual - specific loading vector. - **Neural Network**: Input the individual - specific features into the neural network to predict multi - dimensional clinical information. - **Joint Optimization**: Combine gradient information to simultaneously estimate the underlying network, individual loading vectors and neural network weights. ### Experimental Verification - **Data Set**: Use rs - fMRI data of 52 high - functioning autistic children. - **Clinical Indicators**: Include ADOS (social communication deficits and repetitive behaviors), SRS (social dysfunction) and Praxis (motor skills). - **Performance Evaluation**: Evaluate the multi - score regression performance of the model through ten - fold cross - validation, using median absolute error (MAE) and mutual information (MI) as evaluation indicators. ### Main Contributions - **Innovative Optimization Framework**: For the first time, combine the representation ability of traditional optimization methods and modern neural networks to achieve the modeling and prediction of multi - dimensional clinical features. - **Better Generalization Ability**: Compared with the baseline method, this framework shows better performance in predicting multiple clinical indicators, especially on SRS and Praxis indicators. - **Biological Explanation**: The sub - networks extracted by dictionary learning provide a biological explanation for autism - related brain regions, which is helpful for understanding the mechanisms of neurodevelopmental disorders. In conclusion, this paper successfully solves the challenge of predicting multi - dimensional clinical features from functional connectomics data by proposing a new optimization framework, providing new tools and methods for the research of neurodevelopmental disorders such as autism.

Integrating Neural Networks and Dictionary Learning for Multidimensional Clinical Characterizations from Functional Connectomics Data

Deep sr-DDL: Deep structurally regularized dynamic dictionary learning to integrate multimodal and dynamic functional connectomics data for multidimensional clinical characterizations

A Joint Network Optimization Framework to Predict Clinical Severity from Resting State Functional MRI Data

A Deep-Generative Hybrid Model to Integrate Multimodal and Dynamic Connectivity for Predicting Spectrum-Level Deficits in Autism

Multi-modal deep learning of functional and structural neuroimaging and genomic data to predict mental illness

A generative-discriminative framework that integrates imaging, genetic, and diagnosis into coupled low dimensional space

Deep Multimodal Predictome for Studying Mental Disorders

A Method of Feature Selection via Deep Convolution Neural Networks For Encoding Nonlinear Functional Network Connectivity and Its Application To The Classification of Mental Disorders

Combining Neuroimaging and Omics Datasets for Disease Classification Using Graph Neural Networks

Multi-graph Fusion for Functional Neuroimaging Biomarker Detection

Enhancing Autism Spectrum Disorder identification in multi-site MRI imaging: A multi-head cross-attention and multi-context approach for addressing variability in un-harmonized data

Deep Multimodal Learning for the Diagnosis of Autism Spectrum Disorder

Bayesian Models of Functional Connectomics and Behavior

IFC-GNN: Combining Interactions of Functional Connectivity with Multimodal Graph Neural Networks for ASD Brain Disorder Analysis

Deep learning based joint fusion approach to exploit anatomical and functional brain information in autism spectrum disorders

Feature Learning and Fusion of Multimodality Neuroimaging and Genetic Data for Multi-status Dementia Diagnosis

Multi-task Dictionary Learning based Convolutional Neural Network for Computer aided Diagnosis with Longitudinal Images

Enhancing the Representation of Functional Connectivity Networks by Fusing Multi-View Information for Autism Spectrum Disorder Diagnosis

Multimodal Brain Disease Classification with Functional Interaction Learning from Single fMRI Volume

Neuropsychiatric Disease Classification Using Functional Connectomics -- Results of the Connectomics in NeuroImaging Transfer Learning Challenge

Mapping Multi-Modal Brain Connectome for Brain Disorder Diagnosis Via Cross-Modal Mutual Learning.