Applications of Federated Learning in Mobile Health: Scoping Review

Cross-device federated unsupervised learning for the detection of anomalies in single-lead electrocardiogram signals

BACKGROUND

Federated unsupervised learning offers a promising approach to leveraging decentralized data stored on consumer devices, addressing concerns about privacy and lack of annotation. Single-lead electrocardiograms (ECGs) captured on consumer devices are of particular interest due to the global prevalence of cardiovascular disease. The combination of federated and unsupervised learning on biomedical data in a cross-device environment raises questions regarding feasibility and accuracy, especially when considering heterogeneous data.

METHODS

A randomly selected subset of the Icentia11k open-source dataset containing mobile ECG recordings was used for this study. Heartbeats are labeled as normal, unknown or the pathological classes: premature atrial contraction and premature ventricular contraction. A linear autoencoder model was used as a method to predict the pathological cases using the embedding space and reconstruction error. The model was integrated into a mobile application that supports ECG data recording, preprocessing into heartbeat segments, and participation in a federated learning pipeline as a client node. The autoencoder was trained collaboratively using federated learning with twenty mobile devices, followed by an additional ten epochs of on-device fine-tuning to account for personalization.

RESULTS

The approach yielded a sensitivity of 0.87 and a specificity of 0.8 when the predicted anomalies were compared with the ground truth in a binary fashion. Specifically, the detection rate for premature ventricular contraction was excellent with a sensitivity of 0.97.

CONCLUSION

Overall, the approach proved to be feasible in implementation and competitive in accuracy, specifically when the model was fine-tuned to the subject's data.

How are Machine Learning and Artificial Intelligence Used in Digital Behavior Change Interventions? A Scoping Review

To assess the current real-world applications of machine learning (ML) and artificial intelligence (AI) as functionality of digital behavior change interventions (DBCIs) that influence patient or consumer health behaviors. A scoping review was done across the EMBASE, PsycInfo, PsycNet, PubMed, and Web of Science databases using search terms related to ML/AI, behavioral science, and digital health to find live DBCIs using ML or AI to influence real-world health behaviors in patients or consumers. A total of 32 articles met inclusion criteria. Evidence regarding behavioral domains, target real-world behaviors, and type and purpose of ML and AI used were extracted. The types and quality of research evaluations done on the DBCIs and limitations of the research were also reviewed. Research occurred between October 9, 2023, and January 20, 2024. Twenty-three DBCIs used AI to influence real-world health behaviors. Most common domains were cardiometabolic health (n=5, 21.7%) and lifestyle interventions (n=4, 17.4%). The most common types of ML and AI used were classical ML algorithms (n=10, 43.5%), reinforcement learning (n=8, 34.8%), natural language understanding (n=8, 34.8%), and conversational AI (n=5, 21.7%). Evidence was generally positive, but had limitations such as inability to detect causation, low generalizability, or insufficient study duration to understand long-term outcomes. Despite evidence gaps related to the novelty of the technology, research supports the promise of using AI in DBCIs to manage complex input data and offer personalized, contextualized support for people changing real-world behaviors. Key opportunities are standardizing terminology and improving understanding of what ML and AI are.

Analyzing the Impact of Personalization on Fairness in Federated Learning for Healthcare

As machine learning (ML) usage becomes more popular in the healthcare sector, there are also increasing concerns about potential biases and risks such as privacy. One countermeasure is to use federated learning (FL) to support collaborative learning without the need for patient data sharing across different organizations. However, the inherent heterogeneity of data distributions among participating FL parties poses challenges for exploring group fairness in FL. While personalization within FL can handle performance degradation caused by data heterogeneity, its influence on group fairness is not fully investigated. Therefore, the primary focus of this study is to rigorously assess the impact of personalized FL on group fairness in the healthcare domain, offering a comprehensive understanding of how personalized FL affects group fairness in clinical outcomes. We conduct an empirical analysis using two prominent real-world Electronic Health Records (EHR) datasets, namely eICU and MIMIC-IV. Our methodology involves a thorough comparison between personalized FL and two baselines: standalone training, where models are developed independently without FL collaboration, and standard FL, which aims to learn a global model via the FedAvg algorithm. We adopt Ditto as our personalized FL approach, which enables each client in FL to develop its own personalized model through multi-task learning. Our assessment is achieved through a series of evaluations, comparing the predictive performance (i.e., AUROC and AUPRC) and fairness gaps (i.e., EOPP, EOD, and DP) of these methods. Personalized FL demonstrates superior predictive accuracy and fairness over standalone training across both datasets. Nevertheless, in comparison with standard FL, personalized FL shows improved predictive accuracy but does not consistently offer better fairness outcomes. For instance, in the 24-h in-hospital mortality prediction task, personalized FL achieves an average EOD of 27.4% across racial groups in the eICU dataset and 47.8% in MIMIC-IV. In comparison, standard FL records a better EOD of 26.2% for eICU and 42.0% for MIMIC-IV, while standalone training yields significantly worse EOD of 69.4% and 54.7% on these datasets, respectively. Our analysis reveals that personalized FL has the potential to enhance fairness in comparison to standalone training, yet it does not consistently ensure fairness improvements compared to standard FL. Our findings also show that while personalization can improve fairness for more biased hospitals (i.e., hospitals having larger fairness gaps in standalone training), it can exacerbate fairness issues for less biased ones. These insights suggest that the integration of personalized FL with additional strategic designs could be key to simultaneously boosting prediction accuracy and reducing fairness disparities. The findings and opportunities outlined in this paper can inform the research agenda for future studies, to overcome the limitations and further advance health equity research.

Gait Impairment Analysis Using Silhouette Sinogram Signals and Assisted Knowledge Learning

The analysis of body motion is a valuable tool in the assessment and diagnosis of gait impairments, particularly those related to neurological disorders. In this study, we propose a novel automated system leveraging artificial intelligence for efficiently analyzing gait impairment from video-recorded images. The proposed methodology encompasses three key aspects. First, we generate a novel one-dimensional representation of each silhouette image, termed a silhouette sinogram, by computing the distance and angle between the centroid and each detected boundary points. This process enables us to effectively utilize relative variations in motion at different angles to detect gait patterns. Second, a one-dimensional convolutional neural network (1D CNN) model is developed and trained by incorporating the consecutive silhouette sinogram signals of silhouette frames to capture spatiotemporal information via assisted knowledge learning. This process allows the network to capture a broader context and temporal dependencies within the gait cycle, enabling a more accurate diagnosis of gait abnormalities. This study conducts training and an evaluation utilizing the publicly accessible INIT GAIT database. Finally, two evaluation schemes are employed: one leveraging individual silhouette frames and the other operating at the subject level, utilizing a majority voting technique. The outcomes of the proposed method showed superior enhancements in gait impairment recognition, with overall F1-scores of 100%, 90.62%, and 77.32% when evaluated based on sinogram signals, and 100%, 100%, and 83.33% when evaluated based on the subject level, for cases involving two, four, and six gait abnormalities, respectively. In conclusion, by comparing the observed locomotor function to a conventional gait pattern often seen in healthy individuals, the recommended approach allows for a quantitative and non-invasive evaluation of locomotion.

Tri-branch feature pyramid network based on federated particle swarm optimization for polyp segmentation

Deep learning technology has shown considerable potential in various domains. However, due to privacy issues associated with medical data, legal and ethical constraints often result in smaller datasets. The limitations of smaller datasets hinder the applicability of deep learning technology in the field of medical image processing. To address this challenge, we proposed the Federated Particle Swarm Optimization algorithm, which is designed to increase the efficiency of decentralized data utilization in federated learning and to protect privacy in model training. To stabilize the federated learning process, we introduced Tri-branch feature pyramid network (TFPNet), a multi-branch structure model. TFPNet mitigates instability during the aggregation model deployment and ensures fast convergence through its multi-branch structure. We conducted experiments on four different public datasets：CVC-ClinicDB, Kvasir, CVC-ColonDB and ETIS-LaribPolypDB. The experimental results show that the Federated Particle Swarm Optimization algorithm outperforms single dataset training and the Federated Averaging algorithm when using independent scattered data, and TFPNet converges faster and achieves superior segmentation accuracy compared to other models.

Architectural Design of a Blockchain-Enabled, Federated Learning Platform for Algorithmic Fairness in Predictive Health Care: Design Science Study

BACKGROUND

Developing effective and generalizable predictive models is critical for disease prediction and clinical decision-making, often requiring diverse samples to mitigate population bias and address algorithmic fairness. However, a major challenge is to retrieve learning models across multiple institutions without bringing in local biases and inequity, while preserving individual patients' privacy at each site.

OBJECTIVE

This study aims to understand the issues of bias and fairness in the machine learning process used in the predictive health care domain. We proposed a software architecture that integrates federated learning and blockchain to improve fairness, while maintaining acceptable prediction accuracy and minimizing overhead costs.

METHODS

We improved existing federated learning platforms by integrating blockchain through an iterative design approach. We used the design science research method, which involves 2 design cycles (federated learning for bias mitigation and decentralized architecture). The design involves a bias-mitigation process within the blockchain-empowered federated learning framework based on a novel architecture. Under this architecture, multiple medical institutions can jointly train predictive models using their privacy-protected data effectively and efficiently and ultimately achieve fairness in decision-making in the health care domain.

RESULTS

We designed and implemented our solution using the Aplos smart contract, microservices, Rahasak blockchain, and Apache Cassandra-based distributed storage. By conducting 20,000 local model training iterations and 1000 federated model training iterations across 5 simulated medical centers as peers in the Rahasak blockchain network, we demonstrated how our solution with an improved fairness mechanism can enhance the accuracy of predictive diagnosis.

CONCLUSIONS

Our study identified the technical challenges of prediction biases faced by existing predictive models in the health care domain. To overcome these challenges, we presented an innovative design solution using federated learning and blockchain, along with the adoption of a unique distributed architecture for a fairness-aware system. We have illustrated how this design can address privacy, security, prediction accuracy, and scalability challenges, ultimately improving fairness and equity in the predictive health care domain.