Editorial: Crosstalk between computational medicine and neuroscience in healthcare

Zhijin Wang, Philippe Fournier‐Viger, Yu‐Ping Wang, Yong-Gang FuFrontiers in Neuroscience 2023 · DOI · OpenAlex

Abstract

Modern healthcare is transitioning from reactive symptom management to proactive, individualized care. This shift is driven by the integration of computational intelligence with neuroscience, a synergy that addresses the "data-rich but insight-poor" paradox of contemporary clinical practice. This editorial reviews four pioneering studies that leverage advanced architectures—including Deep Forests, Graph Neural Networks (GNNs), and transfer learning—to enhance protein classification, diabetes management, and early disease detection.

Introduction

While the volume of neurological and clinical data has reached unprecedented levels, traditional medical approaches often struggle to interpret the complex, dynamic relationships between biomarkers and disease progression. To move beyond static expert knowledge, a deeper integration of Artificial Intelligence (AI) into clinical workflows is required. This "crosstalk" between computational medicine and neuroscience is essential for refining clinical decisions and enabling early intervention.

Innovative Methodologies in Applied Intelligence

The research featured in this topic introduces several novel methodologies tailored to specific medical challenges:

Deep Forest for Proteomics

The Golgi_DF model, proposed by Bao et al. (2023), utilizes a "deep forest" algorithm as an alternative to standard deep neural networks. By incorporating Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) layers for robust feature fusion, the model effectively addresses sample imbalance in Golgi protein classification, significantly improving predictive accuracy.

Graph Neural Networks for Diabetes Management

Addressing the heterogeneity of blood glucose data, Huang et al. (2023) introduced the HETER model. By utilizing subsequence repetition and Graph Neural Networks (GNNs) to process variable-length Continuous Glucose Monitoring (CGM) data, the model provides robust and generalizable predictions across diverse patient datasets, outperforming traditional temporal analysis methods.

Medical Ambient Intelligence (MAIS)

Yang et al. (2023) designed a Medical Ambient Intelligence System (MAIS) for the early detection of atherosclerosis. This system integrates the entire clinical process—from data collection to decision visualization. Combined with the "SEC model," MAIS identified Vitamin B12 as a critical biomarker for early-stage detection, demonstrating the potential of smart diagnosis in clinical environments.

Two-Phase Transfer Learning for Autism Screening

To improve accessibility, Li et al. (2023) developed a mobile-friendly facial recognition algorithm for autism screening in children. By employing a two-phase transfer learning approach and multiple classifiers, the researchers created a resource-efficient model capable of high-accuracy screening on consumer-grade devices.

Key Insights and Implications

The collective findings of these studies offer several critical insights for the future of healthcare:

  • Architectural Diversity: The success of the Deep Forest architecture in proteomics suggests that ensemble-based deep learning can be superior to traditional backpropagation models in specific biological contexts.
  • Ambient Diagnostics: The shift toward MAIS indicates a future where the clinical environment itself serves as a diagnostic tool, moving away from episodic testing.
  • Handling Heterogeneity: The use of GNNs to manage patient-specific data variability is a major step toward generalizable medical AI.
  • Democratization of Care: The focus on resource-efficient, mobile-friendly models ensures that advanced diagnostics are accessible for early intervention in diverse settings.

Conclusion and Future Directions

The advancements presented here underscore the transformative potential of computational medicine. However, the transition from research to large-scale clinical validation remains a primary challenge. Future efforts must focus on bridging the multidisciplinary gap between computer scientists and clinicians to build a unified, scalable healthcare infrastructure. The editors hope this research topic serves as a wellspring of inspiration for advancing applied intelligence in the health sector.

Summary updated

Abstract

All submissions were subject to peer review by at least two researchers in the relevant domains. The accepted articles in this Research Topic are briefly introduced below. The paper entitled "Golgi DF: Golgi proteins classification with deep forest" by Bao et al. (2023) presented the Gogli DF model for classifying Golgi proteins. This approach utilizes a deep forest algorithm enhanced by advanced feature extraction and sample balancing techniques. It employs dimensionality reduction, convolutional, and LSTM networks for robust feature fusion. The approach signifies a substantial improvement in predictive accuracy for Golgi protein classification. The paper, titled "Heterogeneous temporal representation for diabetic blood glucose prediction" by Huang et al. (2023) proposed a predictive model HETER for blood glucose levels in multiple patients with diabetes that addresses data heterogeneity and uncertainty. The model uses subsequence repetition and graph neural networks to effectively process and analyze variable-length CGM data sequences and outperforms traditional methods, providing robust and generalizable predictions across diverse patient datasets. Experimental results show the effectiveness of the proposed method. The paper titled "Early prediction of atherosclerosis diagnosis with medical ambient intelligence" by Yang et al. (2023) designed a Medical Ambient Intelligence System (MAIS) for the early detection of atherosclerosis. The system integrates the entire clinical process from data collection to decision visualization, enhancing the diagnosis task's performance. Moreover, the MAIS and SEC model together demonstrate superior performance compared to existing methods and identify Vitamin B12 as a key biomarker for early-stage atherosclerosis detection. This also shows the promising of the medical healthcare system in integrating smart diagnosis techniques. The paper "A face image classification method of autistic children based on the two-phase transfer learning" by Li et al. (2023) proposed an improved facial recognition algorithm to screen for autism in children using mobile-friendly deep learning models. By integrating two transfer learning phases and multiple classifiers, the model aims to increase detection accuracy. This approach offers a quick, resource-efficient method for early autism identification. Editors hope that this research topic will serve as a wellspring of both inspiration and substantive insights, contributing to advancing applied intelligence within the health and healthcare sectors across theoretical and practical dimensions. This Research Topic provides numerous opportunities for researchers and practitioners to further utilize and expand upon the presented research. The editors sincerely appreciate the effort and dedication of all contributors to this research topic.
BibTeX
@inproceedings{wang2023editorial,
  author = {Zhijin Wang and Philippe Fournier‐Viger and Yu‐Ping Wang and Yong-Gang Fu},
  title = {{Editorial: Crosstalk between computational medicine and neuroscience in healthcare}},
  booktitle = {Frontiers in Neuroscience},
  year = {2023},
  doi = {10.3389/fnins.2023.1333227},
}
Figure