GAN-Based Semi-Supervised Learning Approach for Clinical Decision Support in Health-IoT Platform

Yun Yang, Fengtao Nan, Po Yang, Qiang Meng, Yingfu Xie, Dehai Zhang, Khan MuhammadIEEE Access 2019 · OpenReview

Abstract

With the development of the Internet of Things (IoT) technology, its application in the medical field becomes more and more extensive. However, with a dramatic increase in medical data obtained from the IoT-based health service system, labeling a large number of medical data requires high cost and relevant domain knowledge. Therefore, how to use a small number of labeled medical data reasonably to build an efficient and high-quality clinical decision support model in the IoT-based platform has been an urgent research topic. In this paper, we propose a novel semi-supervised learning approach in association with generative adversarial networks (GANs) for supporting clinical decision making in the IoT-based health service system. In our approach, GAN is adopted to not only increase the number of labeled data but also to compensate the imbalanced labeled classes with additional artificial data in order to improve the semi-supervised learning performance. Extensive evaluations on a collection of benchmarks and real-world medical datasets show that the proposed technique outperforms the others and provides a potential solution for practical applications.

Introduction

The rise of the Internet of Things (IoT) in healthcare (Health-IoT) has led to an explosion of medical data from wearables and remote monitoring. However, this data is often "dark"—it lacks the expert labels (diagnoses) required to train standard machine learning models. Manually labeling this data is prohibitively expensive and requires significant time from specialized clinicians.

Furthermore, medical datasets are notoriously imbalanced; cases of specific diseases (the minority class) are much rarer than healthy samples (the majority class). Traditional methods like SMOTE often struggle with noise or disrupting class boundaries. This research addresses these challenges by creating a system that can learn effectively from a small "seed" of labeled data while handling the inherent class imbalances of real-world clinical settings.

Methodology

The proposed framework integrates semi-supervised learning (SSL) with Generative Adversarial Networks (GANs) across four main modules.

The proposed Health-IoT clinical decision support framework.
  1. Data Acquisition & Processing: Raw IoT data is collected and undergoes feature selection to identify relevant clinical indicators.
  2. GAN-Based Data Augmentation: A GAN is trained to capture the underlying probability distribution of the original dataset. It generates "artificial" labeled samples to both increase the total training volume and specifically "pad" the minority classes to achieve a balanced distribution.
  3. Hybrid Semi-Supervised Learning: This stage combines self-training and co-training. Two different "base learners" (e.g., SVM and KNN) are trained on the augmented labeled set. These learners then predict labels for the unlabeled data.
  4. The Agreement Filter: Only unlabeled instances where both learners agree on the pseudo-label are added to the training set for the next iteration. This significantly reduces the risk of "label noise" that usually plagues semi-supervised methods.
Algorithm for GAN-based semi-supervised learning.

Experimental Setup and Datasets

The researchers validated their approach using two distinct sources:

  • UCI Machine Learning Repository: 20 medical benchmark datasets (including Breast Cancer, Heart Disease, and Diabetes). Ten were chosen for being balanced and ten for being imbalanced to test the GAN's oversampling capabilities.
  • Cerebral Stroke Dataset: A real-world dataset from a hospital in Kunming, China. It contains 11,039 patient records with 33 attributes. Crucially, only 100 instances were initially labeled, representing a classic "small-labeled/imbalanced" challenge.

Key Results and Insights

The proposed method was compared against state-of-the-art SSL algorithms like S3VM, S4VM, and TSVM.

  • Accuracy Boost: On the real-world stroke dataset, the GAN-based approach outperformed competitors by 2% to 4% in classification accuracy.
  • Robustness to Imbalance: On the 10 imbalanced UCI datasets, the proposed method achieved the best results in 8 out of 10 cases using a linear kernel.
  • Stability through Committee: By requiring two different algorithms to agree on a pseudo-label before accepting it, the system prevents a single biased classifier from "hallucinating" incorrect labels.
  • Superior Sampling: Unlike traditional interpolation-based oversampling, GANs learn the underlying probability distribution of the data, allowing them to generate more realistic "artificial" patients.

Conclusion

The primary trade-off identified is computational overhead, as training GANs and running iterative pseudo-labeling is resource-intensive. However, the approach provides a potential solution for practical applications where labeled data is scarce. Future work will investigate which specific combinations of "base learners" provide the most reliable "agreement filter" for different types of medical data.

Summary updated

Abstract

With the development of the Internet of Things (IoT) technology, its application in the medical field becomes more and more extensive. However, with a dramatic increase in medical data obtained from the IoT-based health service system, labeling a large number of medical data requires high cost and relevant domain knowledge. Therefore, how to use a small number of labeled medical data reasonably to build an efficient and high-quality clinical decision support model in the IoT-based platform has been an urgent research topic. In this paper, we propose a novel semi-supervised learning approach in association with generative adversarial networks (GANs) for supporting clinical decision making in the IoT-based health service system. In our approach, GAN is adopted to not only increase the number of labeled data but also to compensate the imbalanced labeled classes with additional artificial data in order to improve the semi-supervised learning performance. Extensive evaluations on a collection of benchmarks and real-world medical datasets show that the proposed technique outperforms the others and provides a potential solution for practical applications.
BibTeX
@inproceedings{yang2019gan,
  author = {Yun Yang and Fengtao Nan and Po Yang and Qiang Meng and Yingfu Xie and Dehai Zhang and Khan Muhammad},
  title = {{GAN-Based Semi-Supervised Learning Approach for Clinical Decision Support in Health-IoT Platform}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=xH7CENJKwo8},
}
Figure