Scholarlink

The emergence of new and re-emerging viral diseases (i.e., COVID-19, Influenza, and others, including Monkeypox and Dengue) underscores the urgent need for fast, easy-to-understand, and reliable diagnostic methods in health care. It is the challenge of those issues that motivated the development of NextGenViDx. Consequently, the aim of this report is to describe NextGenViDx - a multi-sourced and multi-modal artificial intelligence (AI) platform for facilitating safer and more transparent processes of identifying diseases. NextGenViDx differs from other approaches in its use of many different clinical data sources (e.g., demographics, symptom descriptions, CT scans and GAN-generated synthetic signals) instead of a single source of input or a single centralized database. A distributed, federated learning design permits models to be trained locally and therefore ensures that sensitive patient data remain on-premises at the originating institution. The main element of the system is a lightweight multilayer perceptron (MLP) classifier that was chosen because of the speed of evaluation and compatibility with federated learning designs. NextGenViDx was trained using a modest-sized synthetic dataset (n = 20 cases) that included six different viral diseases. Despite being trained with a very limited size dataset, NextGenViDx exhibited excellent learning characteristics and reliability in terms of its classification performance. In addition, visual representations such as confusion matrices and ROC curves allow clinicians to gain insight into the reasoning process behind each prediction. NextGenViDx demonstrated high levels of accuracy for all six disease categories (approximately 90%) and demonstrated a good balance of precision and recall. Furthermore, the macro area under the receiver operating characteristic curve (AUC) of .93 of NextGenViDx demonstrates strong discriminatory capabilities to distinguish between viral infections that share similar symptoms. Finally, due to its low computational requirements, NextGenViDx can quickly evaluate new cases and complete training iterations in a few milliseconds, which makes it ideal for supporting real-time clinical work flows in a wide range of practical health care environments. Moreover, due to its decentralized nature, NextGenViDx will not require the collection of patients’ data centrally and it has a scalable architecture that supports the integration of additional modalities. .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Federated and Explainable Vision-Transformer Framework Enhanced with Generative AI for Real-Time, Accurate, and Ethical Viral Disease Detection Across Diverse Clinical Modalities

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Asadi Srinivasulu,
  • Anupam Agrawal

摘要

The emergence of new and re-emerging viral diseases (i.e., COVID-19, Influenza, and others, including Monkeypox and Dengue) underscores the urgent need for fast, easy-to-understand, and reliable diagnostic methods in health care. It is the challenge of those issues that motivated the development of NextGenViDx. Consequently, the aim of this report is to describe NextGenViDx - a multi-sourced and multi-modal artificial intelligence (AI) platform for facilitating safer and more transparent processes of identifying diseases. NextGenViDx differs from other approaches in its use of many different clinical data sources (e.g., demographics, symptom descriptions, CT scans and GAN-generated synthetic signals) instead of a single source of input or a single centralized database. A distributed, federated learning design permits models to be trained locally and therefore ensures that sensitive patient data remain on-premises at the originating institution. The main element of the system is a lightweight multilayer perceptron (MLP) classifier that was chosen because of the speed of evaluation and compatibility with federated learning designs. NextGenViDx was trained using a modest-sized synthetic dataset (n = 20 cases) that included six different viral diseases. Despite being trained with a very limited size dataset, NextGenViDx exhibited excellent learning characteristics and reliability in terms of its classification performance. In addition, visual representations such as confusion matrices and ROC curves allow clinicians to gain insight into the reasoning process behind each prediction. NextGenViDx demonstrated high levels of accuracy for all six disease categories (approximately 90%) and demonstrated a good balance of precision and recall. Furthermore, the macro area under the receiver operating characteristic curve (AUC) of .93 of NextGenViDx demonstrates strong discriminatory capabilities to distinguish between viral infections that share similar symptoms. Finally, due to its low computational requirements, NextGenViDx can quickly evaluate new cases and complete training iterations in a few milliseconds, which makes it ideal for supporting real-time clinical work flows in a wide range of practical health care environments. Moreover, due to its decentralized nature, NextGenViDx will not require the collection of patients’ data centrally and it has a scalable architecture that supports the integration of additional modalities. .