EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

A Cloud Native Data and Intelligence Framework for Medical Reasoning and Digital Health Analysis

Pushkar Arya, Beena M, Manju Khanna

Procedia Computer Science · 2026 · 283 · 2877-2894

DOI: 10.1016/j.procs.2026.06.356 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Procedia Computer Science, 2026. Verfasst von Pushkar Arya und 2 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The generic large language models (LLMs) mainly rely on the exposed data during pre-training to generate a response. They generally face difficulty generalizing domain-specific problems due to their limited capability regarding semantic search and analytical reasoning over vast subject-specific data. To obtain reliable responses in the critical medical domain of disease investigation, we propose a Cloud Native Data and Intelligence Framework for Medical Reasoning and Digital Health Analysis (DHA) with the ability to hold conversations with in-depth business expertise and link generic LLMs to medical knowledge bases. This hybrid data intelligence framework provides multi-stage collaborative inference achieved through extraction, categorization, and LLM Re-Act Analysis in conjunction with scalable, resilient, performant execution of these multi-stage tasks over the AWS cloud-native technologies and evaluated against seven standard LLM techniques of Prompt engineering, Retrieval Augmented Generation (RAG), and two advanced metrics of Semantic accuracy (SA) and Contextual accuracy (CA). The proposed framework can respond to any medical reasoning and decision-making requirements for complex real-world medical diagnoses and consultation use cases through a collaborative decision-making mechanism of LLM agents, which conducts preliminary medical complexity bifurcation and decides the allocation of individual specialists or a cross-disciplinary mixture of domain experts to address the situation. To evaluate and benchmark the accuracy and performance of the DHA system, we conduct experiments on a sizable number of Infused medical data created with an amalgamation of 7 Public and 1 Private dataset pitted against 5 Pretrained Models LLAMA, BERT, GPT 3.5, GPT4, Vicuna achieving the accuracy of 0.65%, 0.58%, 0.44%, 0.51%, and 0.69% on baseline methods of prompt shots, chain-of-thoughts, voting, debate, reconciliation, and adaptive conversations. While DHA, equipped with multi-stages inference and advanced criteria of SA, CA outperforms them on all Datasets with the highest accuracy of 0.89%, 0.87%, 0.90%, 0.86%, and 0.84% on baseline methods and an advance accuracy 0.93% and 94% on SA, CA methods, respectively, and has the immense potential to improve disease prediction and individualized health care significantly.

Zitieren

APA

Arya, P., M, B., & Khanna, M. (2026). A Cloud Native Data and Intelligence Framework for Medical Reasoning and Digital Health Analysis. Procedia Computer Science, 283, 2877-2894. https://doi.org/10.1016/j.procs.2026.06.356

Vancouver

Arya P, M B, Khanna M. A Cloud Native Data and Intelligence Framework for Medical Reasoning and Digital Health Analysis. Procedia Computer Science. 2026;283:2877-2894. doi:10.1016/j.procs.2026.06.356

BibTeX
@article{Arya2026,
  author = {Arya, Pushkar and M, Beena and Khanna, Manju},
  title = {A Cloud Native Data and Intelligence Framework for Medical Reasoning and Digital Health Analysis},
  journal = {Procedia Computer Science},
  year = {2026},
  volume = {283},
  pages = {2877--2894},
  doi = {10.1016/j.procs.2026.06.356}
}
Kostenlos starten →