SEARCHNARRATIVE_REVIEWdigital health, machine learning

narrative_review·digital health, machine learning, transfer learning, wearable sensors, medical imaging·PMC10744730

Survey of Transfer Learning Approaches in the Machine Learning of Digital Health Sensing Data

Journal of Personalized Medicine · 2 authors, 1 centre

AI SUMMARY

FIDELITY 80%

This summary was generated by AI from a single paper. It has not been reviewed by a clinician and is not clinical advice. Verify against the source before acting on it.

AGENT 01 · STRUCTURED SUMMARY

TRACE TO SOURCE

“

TL;DR

This survey comprehensively reviews transfer learning (TL) methods applied to digital health sensing data to overcome challenges like limited data, privacy, and computational demands. It categorizes TL techniques—including fine-tuning, feature extraction, domain adaptation, multitask learning, federated learning, and few-shot learning—and demonstrates their effectiveness in improving diagnostic accuracy and efficiency across various healthcare applications. The paper provides a structured guide for researchers to select and apply appropriate TL strategies for enhancing digital health technologies.

Full summary

4,011 CHARS

**Background:** Digital health (DH) leverages information and communication technologies to enhance healthcare services, utilizing portable and nonportable sensing technologies such as wearable devices, biosensors, and medical imaging. Machine learning (ML) and deep learning (DL) are central to analyzing DH data for diagnosis, prognosis, and personalized care. However, developing accurate ML models from scratch is challenging due to limited data availability, data quality issues, privacy concerns, and high computational costs. Transfer learning (TL) has emerged as a solution by transferring knowledge from a related source task to improve performance on a target task, reducing the need for large labeled datasets and extensive computation.

**Methods:** This survey paper systematically reviews TL approaches applied to DH sensing data. The authors sourced literature from databases including IEEE Xplore, MDPI, PubMed, Google Scholar, and ACM Digital Library, using keywords related to TL and DH sensing technologies. The paper first categorizes DH sensing technologies into portable (e.g., wearable, implantable, ingestible, smartphone-based) and nonportable (e.g., stationary imaging, environmental sensors). It then explains TL concepts, including domains, tasks, and categories (inductive, transductive, unsupervised). The core of the review examines six TL methods: (1) feature extraction, (2) fine-tuning (partial, full, progressive, adaptive), (3) domain adaptation, (4) multitask learning, (5) few-/one-/zero-shot learning, and (6) federated learning. For each method, the paper discusses strategies, applications, and key studies, providing detailed examples from DH domains such as human activity recognition, sleep staging, cardiac monitoring, medical imaging, and emotion recognition.

**Key Results:** The survey analyzes 27 studies in detail. Key findings include: - Feature extraction using pretrained DL models (e.g., ResNet50, DenseNet) achieved high accuracy in COVID-19 detection from X-rays (99.86% with SVM) and ECG arrhythmia classification (97.23% accuracy). - Fine-tuning improved human activity recognition accuracy by up to 43% and reduced training time by 66% (study [36]). In sleep staging, fine-tuning a CRNN on clinical PSG data achieved a Cohen's kappa of 0.78, close to expert level. - Domain adaptation methods, such as stratified transfer learning, improved activity recognition accuracy by 7.68% over state-of-the-art methods. - Multitask learning jointly performing brain tumor classification and segmentation achieved 98.04% classification accuracy and a Dice coefficient of 0.74 for segmentation. - Few-shot learning for arrhythmia detection outperformed other meta-learning methods by 3–12% in accuracy for 1-shot 2-way tasks. - Federated learning frameworks (e.g., FedHealth) improved human activity recognition accuracy by 21.6% and 16.8% over non-federated systems while preserving privacy. - Combined domain and decision transfer for sleep staging from PPG data achieved a Cohen's kappa of 0.65 ± 0.11 and accuracy of 76.36 ± 7.57%.

**Clinical Implications:** TL methods significantly enhance the accuracy, efficiency, and privacy of ML models in digital health, enabling robust diagnosis and monitoring even with limited data. These approaches facilitate early disease detection (e.g., COVID-19, breast cancer, arrhythmias), personalized treatment (e.g., glucose monitoring, sleep staging), and remote patient monitoring (e.g., fall detection, activity recognition). Federated learning addresses critical privacy concerns, allowing collaborative model training across institutions without sharing sensitive patient data. The survey underscores that selecting appropriate TL strategies based on data characteristics and task requirements is crucial for maximizing clinical benefit. Future research should focus on adaptive learning for real-time data, enabling TL on edge devices, and addressing challenges like negative transfer and model interpretability.

STUDY TYPE

narrative_review

SPECIALTY

digital health

SUMMARISED BY

AI pipeline

FIDELITY CHECK

80% · B

ORIGINAL PAPER

PMC10744730

Survey of Transfer Learning Approaches in the Machine Learning of Digital Health Sensing Data

Journal of Personalized Medicine

READ THE FULL PAPER

JUMP TO SECTION

ABSTRACT

Machine learning and digital health sensing data have led to numerous research achievements aimed at improving digital health technology. However, using machine learning in digital health poses challenges related to data availability, such as incomplete, unstructured, and fragmented data, as well as issues related to data privacy, security, and data format standardization. Furthermore, there is a risk of bias and discrimination in machine learning models. Thus, developing an accurate prediction model from scratch can be an expensive and complicated task that often requires extensive experiments and complex computations. Transfer learning methods have emerged as a feasible solution to address these issues by transferring knowledge from a previously trained task to develop high-performance prediction models for a new task. This survey paper provides a comprehensive study of the effectiveness of transfer learning for digital health applications to enhance the accuracy and efficiency of diagnoses and prognoses, as well as to improve healthcare services. The first part of this survey paper presents and discusses the most common digital health sensing technologies as valuable data resources for machine learning applications, including transfer learning. The second part discusses the meaning of transfer learning, clarifying the categories and types of knowledge transfer. It also explains transfer learning methods and strategies, and their role in addressing the challenges in developing accurate machine learning models, specifically on digital health sensing data. These methods include feature extraction, fine-tuning, domain adaptation, multitask learning, federated learning, and few-/single-/zero-shot learning. This survey paper highlights the key features of each transfer learning method and strategy, and discusses the limitations and challenges of using transfer learning for digital health applications. Overall, this paper is a comprehensive survey of transfer learning methods on digital health sensing data which aims to inspire researchers to gain knowledge of transfer learning approaches and their applications in digital health, enhance the current transfer learning approaches in digital health, develop new transfer learning strategies to overcome the current limitations, and apply them to a variety of digital health technologies.

CITATION EXPORT

Lina Chato, Emma Regentova. Survey of Transfer Learning Approaches in the Machine Learning of Digital Health Sensing Data. Journal of Personalized Medicine. (2023). https://doi.org/10.3390/jpm13121703

AUTHORS

2 · 1 CENTRES

Lina Chato, Emma Regentova

Department of Electrical and Computer Engineering, University of Nevada, Las Vegas, NV 89154, USA; [email protected]

EDITOR / REVIEWER

Pascale Gauthier