一句话：提出模块化框架 TechnoHealth，用混合数据模型统一异构可穿戴数据，支持可复现的精准医疗分析与模型训练。
可穿戴设备产生长期连续的个体生活方式测量，但数据异构、访问困难、标准化与解释障碍阻碍临床研究者有效处理和训练模型，导致研究偏差与模型不可泛化。作者提出 TechnoHealth，一个模块化软件框架，采用结构化元数据与非结构化传感器数据结合的混合数据模型，保留原始信号并支持灵活的模型就绪转换，同时遵循模块化、透明性和可扩展性等软件工程原则。通过试点研究验证端到端流水线，重点评估异构信号摄入及其向节律性检测模型的转换。该框架支持更广泛可扩展的健康研究，促进可复现科学并加深对健康结果驱动因素的理解。
Wearable devices generate continuous, long term measurements of individual lifestyle, offering potential for actionable insights in precision healthcare. However, from a clinical researcher’s perspective, challenges such as data heterogeneity, difficulties in access, standardization, and interpretation hinder effective processing and model training for intelligent sensor data analysis. These challenges are further exacerbated by access restrictions and inconsistent data formats across devices, often leading to bias in studies and ungeneralizable machine learning models. To address these issues, we present TechnoHealth, a modular software framework designed to address the significant challenges of data fragmentation and heterogeneity inherent in utilizing data from diverse wearable devices for precision healthcare research. Our architecture utilizes a hybrid data model that combines structured metadata (for context and provenance) with unstructured sensor data, thereby preserving the raw signal characteristics and enabling flexible, model ready transformations. A pilot study is described to validate the framework’s end to end pipeline, focusing on the ingestion of heterogeneous signals and their transformation for rhythmicity detection models. The platform’s design adheres to software engineering principles, including modularity, for enhanced transparency and extensibility. Technohealth supports broader and more scalable health studies, which in turn can lead to more reproducible science and improved understanding of drivers of health outcomes.
@inproceedings{MirzaseyediLLH26,
title = {Technohealth: A Modular Framework for Reproducible Research in Precision Healthcare with Heterogeneous Wearable Data},
author = {Melika Mirzaseyedi and Paula Lago and Laurie Anne Laberge and Abdelwahab Hamou-Lhadj},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3786581.3786939}
}