In any age population, neurological disorders dramatically impact the lives of patients, their families, and societies. However, as a vulnerable population, children suffer unique forms and consequences of various neurological disorders, such as attention-deficit hyperactivity disorders (ADHD), autism spectrum disorders (ASD), cerebral palsy, epilepsy, and traumatic brain injury. Among these unique challenges are the specific technological needs and constraints in studying neurological disorders in children, compounded by their stage of development, which introduces uncertainty and further complexity. In order to unravel these complexities, indebted to the availability of high-dimensional data and computing capabilities, approaches based on machine learning have rapidly emerged a new trend aimed at better understanding pathways, accurately diagnosing, and better managing these disorders. Deep learning has recently gained an ever-increasing role in the era of health and medical investigations under the umbrella of machine learning-based solutions for the study of neurological disorders.
Nurtured by the scientific exploration of the learning model in the cerebral cortex, deep learning algorithms have rapidly evolved into a diverse group of applications competing with their siblings, the traditional machine learning algorithms. Projects like AlphaGo or TossingBot, whose core is built on deep learning infrastructure, have demonstrated high performance, with evidence of outperforming humans in specific. While deep learning has relieved the burden of systematic feature engineering, it currently suffers from the time needed for training, high cost, and the size of the data. Aside from these, the lack of validated domain-specific neural network architecture and the reported instabilities in the fine-tuning phase of deep learners have led researchers to limit their deployment to a routine based on a guess and check strategy. The guess and check strategy for the design and implementation of deep neural networks needs to be improved when the problem becomes more complex, e.g., including time-varying and context-dependent processes, as is the case for pediatric neurological disorders. The varying dynamics of the problem, which the deep learner solves, is due to the joint nature of two inter-related categories of processes, the ones related to brain development and those led by the neurological disorder, both of which are highly dynamic processes depending on factors like age. Therefore, the design and deployment of deep learners for pediatric neurological disorders are like hitting moving targets. The application of deep neural networks needs to be further enriched with a more objective design and deployment, fitting the problem constraints we have reviewed. When addressed adequately despite challenges, these constraints will open a new era of scientific discoveries and applied solutions to the field.
We, therefore, cordially invite authors to submit their manuscripts (hypothesis & theory, methods, original research, review and mini-review, perspective, general commentary, new discovery) to the current Research Topic on “Deep learning for neurological disorders in children”. Specific themes we would like contributors to address include, but not are limited to:
• Deep learning for the diagnosis of neurological disorders in children
• Prognostic values of deep learning in pediatric neurological disorders
• The role of deep learning in delineating pathways of neurological disorders in children
• Methodological constraints and solutions to the successful deployment of deep learning algorithms for studying neurological disorders in the pediatric population.
• Deep learning as a bridge from the basic science of pediatric neurological disorders to its clinical application.
• Deep learning as a decision guide supported by the multi-modal big data for studying neurological disorders in children.
In any age population, neurological disorders dramatically impact the lives of patients, their families, and societies. However, as a vulnerable population, children suffer unique forms and consequences of various neurological disorders, such as attention-deficit hyperactivity disorders (ADHD), autism spectrum disorders (ASD), cerebral palsy, epilepsy, and traumatic brain injury. Among these unique challenges are the specific technological needs and constraints in studying neurological disorders in children, compounded by their stage of development, which introduces uncertainty and further complexity. In order to unravel these complexities, indebted to the availability of high-dimensional data and computing capabilities, approaches based on machine learning have rapidly emerged a new trend aimed at better understanding pathways, accurately diagnosing, and better managing these disorders. Deep learning has recently gained an ever-increasing role in the era of health and medical investigations under the umbrella of machine learning-based solutions for the study of neurological disorders.
Nurtured by the scientific exploration of the learning model in the cerebral cortex, deep learning algorithms have rapidly evolved into a diverse group of applications competing with their siblings, the traditional machine learning algorithms. Projects like AlphaGo or TossingBot, whose core is built on deep learning infrastructure, have demonstrated high performance, with evidence of outperforming humans in specific. While deep learning has relieved the burden of systematic feature engineering, it currently suffers from the time needed for training, high cost, and the size of the data. Aside from these, the lack of validated domain-specific neural network architecture and the reported instabilities in the fine-tuning phase of deep learners have led researchers to limit their deployment to a routine based on a guess and check strategy. The guess and check strategy for the design and implementation of deep neural networks needs to be improved when the problem becomes more complex, e.g., including time-varying and context-dependent processes, as is the case for pediatric neurological disorders. The varying dynamics of the problem, which the deep learner solves, is due to the joint nature of two inter-related categories of processes, the ones related to brain development and those led by the neurological disorder, both of which are highly dynamic processes depending on factors like age. Therefore, the design and deployment of deep learners for pediatric neurological disorders are like hitting moving targets. The application of deep neural networks needs to be further enriched with a more objective design and deployment, fitting the problem constraints we have reviewed. When addressed adequately despite challenges, these constraints will open a new era of scientific discoveries and applied solutions to the field.
We, therefore, cordially invite authors to submit their manuscripts (hypothesis & theory, methods, original research, review and mini-review, perspective, general commentary, new discovery) to the current Research Topic on “Deep learning for neurological disorders in children”. Specific themes we would like contributors to address include, but not are limited to:
• Deep learning for the diagnosis of neurological disorders in children
• Prognostic values of deep learning in pediatric neurological disorders
• The role of deep learning in delineating pathways of neurological disorders in children
• Methodological constraints and solutions to the successful deployment of deep learning algorithms for studying neurological disorders in the pediatric population.
• Deep learning as a bridge from the basic science of pediatric neurological disorders to its clinical application.
• Deep learning as a decision guide supported by the multi-modal big data for studying neurological disorders in children.