Tyee: A Unified, Modular, and Fully-Integrated Configurable Toolkit for Intelligent Physiological Health Care

Tyee: A Unified, Modular, and Fully-Integrated Configurable Toolkit for Intelligent Physiological Health Care Tao Zhou∗ Hunan University Changsha, China zhooutao@hnu.edu.cn Lingyu Shu∗ Hunan University Changsha, China shulingyu@hnu.edu.cn Zixing Zhang† Hunan University Changsha, China zixingzhang@hnu.edu.cn Jing Han† University of Cambridge Cambridge, UK jh2298@cam.ac.uk Abstract Deep learning has shown great promise in physiological signal anal- ysis, yet its progress is hindered by heterogeneous data formats, inconsistent preprocessing strategies, fragmented model pipelines, and non-reproducible experimental setups. To address these limi- tations, we present Tyee, a unified, modular, and fully-integrated configurable toolkit designed for intelligent physiological health- care. Tyee introduces three key innovations: (1) a unified data interface and configurable preprocessing pipeline for 12 kinds of signal modalities; (2) a modular and extensible architecture en- abling flexible integration and rapid prototyping across tasks; and (3) end-to-end workflow configuration, promoting reproducible and scalable experimentation. Tyee demonstrates consistent prac- tical effectiveness and generalizability, outperforming or match- ing baselines across all evaluated tasks (with state-of-the-art re- sults on 12 of 13 datasets). The Tyee toolkit is released at https: //github.com/SmileHnu/Tyee and actively maintained. CCS Concepts • Software and its engineering →Open source model; • Ap- plied computing →Life and medical sciences. Keywords Open source, physiological signal, health and well-being 1 Introduction Physiological signals, such as electroencephalography (EEG), elec- trocardiography (ECG), and electromyography (EMG), are essential to understand physiological and cognitive states [7, 9]. Their grow- ing utilization in intelligent healthcare has driven deep learning methods for physiological signal analysis [1, 3]. In particular, these methods have shown promising performance in critical healthcare tasks from disease prediction to patient monitoring [5, 12]. Recently, several open-source toolkits have been developed to support physiological signal analysis. As summarized in Table 1, these toolkits provide a strong foundation and have contributed significantly to advancing the field. However, despite their utility, researchers and developers (e.g., physiological signal engineers and machine learning practitioners) continue to face three key challenges that limit broader adoption. First, data integration remains a major challenge due to fragmented data formats and inconsistent prepro- cessing pipelines across datasets; a simple, flexible interface that hides these differences and makes it easier to work with data from many sources in a consistent way is still lacking. Second, model ∗These authors contributed equally to this research. †Zixing Zhang and Jing Han are corresponding authors. development is often slowed by rigid, monolithic toolkits that limit experimentation and make fair comparisons difficult; a modular framework that encapsulates core components while enabling in- dependent configuration and substitution remains an unmet need. Third, inconsistent workflows, hidden defaults, and inflexible code- bases often lead to inconsistent and unreliable outcomes; a stream- lined system that automates the pipeline—reducing manual coding while maintaining transparency—remains critically needed. To address these three challenges, we introduce an open-source Tyee: a unified, modular, and fully-integrated configurable deep learn- ing Toolkit for physiological health care. The name Tyee, pronounced like the Chinese term “太医” (meaning imperial physician), re- flects its mission to bridge intelligent computing with physiological healthcare. As the first toolkit of its kind, Tyee pioneers accessible physiological signal analysis. It is designed to serve both novice users requiring ready-to-use pipelines and advanced researchers and developers focused on developing, evaluating, or benchmarking cus- tom models across diverse datasets and tasks. To accomplish this, it provides: (i) a unified data interface that abstracts dataset-specific preprocessing through configurable pipelines, enabling seamless integration and reuse; (ii) a modular architecture supporting flexible configuration (models, loss functions, optimizers, and evaluation metrics), enabling rapid prototyping and fair benchmarking; and (iii) holistic workflow configuration that automates the entire ex- perimental process, from data preprocessing to evaluation, boosting reproducibility and scalability with minimal coding. 2 Tyee: Design & Features In this section, we first present the overall architecture of Tyee, followed by a detailed discussion of its key features, including the unified data interface, configurable preprocessing, modular component design, and end-to-end configuration. 2.1 Design Principles & Tyee Architecture Inspired by prior works [4, 11] and t

…(Full text truncated)…

This content is AI-processed based on ArXiv data.