Effective Heart Disease Detection Based on Quantitative Computerized Traditional Chinese Medicine Using Representation Based Classifiers

Ting Shu, Bob Zhang, Yuan Yan TangEvidence-based 2017 · DOI

TL;DR

This paper introduces a novel, non-invasive method for detecting heart disease (HD) by analyzing facial key block color features, inspired by Traditional Chinese Medicine. Using a Probabilistic Collaborative Representation Based Classifier (ProCRC) on a new dataset of facial images, the approach achieves the highest accuracy among several tested classifiers, offering a promising alternative to traditional diagnostic methods.

What Problem Does This Paper Address?

Heart disease remains the leading cause of death globally, and current diagnostic methods, such as blood tests, ECGs, and imaging scans, are often time-consuming, invasive, or both. This creates a need for more accessible and less burdensome diagnostic tools. The paper aims to bridge this gap by developing a non-invasive, computerized method for heart disease detection using facial image analysis, drawing upon the principles of Traditional Chinese Medicine (TCM) which posits that facial characteristics can reflect internal organ health.

Methods

The core of the proposed method involves extracting quantitative color features from specific facial regions, termed "key blocks," and then classifying these features using a machine learning model.

  • Facial Key Block Extraction: Four key facial regions (forehead, left cheek, right cheek, and nose bridge) are automatically identified and extracted from facial images. This process is guided by pupil detection and uses a specially designed capture device to minimize environmental variations.
  • Color Feature Extraction: A color gamut with six predefined color centroids is used to analyze the color distribution within each facial key block. For each block, a feature vector is generated representing the proportion of pixels that match each of the six color centroids.
  • Classifier: The Probabilistic Collaborative Representation Based Classifier (ProCRC) is employed for classification. ProCRC is an extension of the Collaborative Representation Based Classifier (CRC), which itself is an improvement over the Sparse Representation Based Classifier (SRC). ProCRC incorporates probabilistic theory to enhance its classification performance.
  • Optimization: The paper also details an analysis to find the optimal parameters (α and β) for the ProCRC classifier to maximize its accuracy for heart disease detection.

Datasets

The study utilized a newly collected dataset comprising:

  • 581 heart disease (HD) samples
  • 581 healthy (H) samples

These samples were collected from the Guangdong Provincial TCM Hospital in 2015. The dataset includes information on gender and age distribution, with some "no record" (NR) cases. Ethical approval was obtained, and all participants provided consent.

Key Insights and Ideas

  • TCM as a Foundation for Non-Invasive Diagnostics: The paper demonstrates that principles from Traditional Chinese Medicine, specifically the idea that facial color can indicate internal health, can be quantitatively translated into effective diagnostic features for modern machine learning.
  • Facial Key Blocks as Informative Features: The research highlights that specific facial regions (forehead, cheeks, nose bridge) contain color information that is discriminative for heart disease detection, even when the differences are not readily apparent to the naked eye.
  • ProCRC's Superiority: The Probabilistic Collaborative Representation Based Classifier (ProCRC) consistently outperformed other traditional and state-of-the-art classifiers, suggesting its robustness and effectiveness for this specific task.
  • Combination of Features is Key: Combining features from multiple facial key blocks (specifically FHB + LCB + NBB) yielded the highest detection accuracy, indicating that a holistic view of facial color patterns is more powerful than analyzing isolated regions.

Key Results

  • The proposed ProCRC method, using a combination of the Forehead Block (FHB), Left Cheek Block (LCB), and Nose Bridge Block (NBB), achieved the highest accuracy of 88.01%.
  • This best result also yielded a sensitivity of 84.95% and a specificity of 91.07%.
  • The ProCRC method outperformed 10 other classifiers, including k-NN, SVM, SRC, DL with SRC, and CRC, by an average of 3.65% in accuracy when using the FHB + LCB + NBB combination.
  • The paper also showed that the ProCRC method can correctly classify facial blocks that are difficult to distinguish by visual inspection alone.

Limitations and Future Work

  • Limitations: The paper acknowledges that while the ProCRC method shows promise, the differences in facial key blocks between HD and healthy individuals can be subtle and not easily discernible by the naked eye. The study also notes that missing gender and age information did not affect their specific analysis, but this could be a factor in broader applications.
  • Future Work: The authors suggest exploring and extracting additional features from the facial key blocks. Furthermore, they propose developing and applying other representation learning algorithms to further enhance HD detection capabilities.

Summary updated

Abstract

At present, heart disease is the number one cause of death worldwide. Traditionally, heart disease is commonly detected using blood tests, electrocardiogram, cardiac computerized tomography scan, cardiac magnetic resonance imaging, and so on. However, these traditional diagnostic methods are time consuming and/or invasive. In this paper, we propose an effective noninvasive computerized method based on facial images to quantitatively detect heart disease. Specifically, facial key block color features are extracted from facial images and analyzed using the Probabilistic Collaborative Representation Based Classifier. The idea of facial key block color analysis is founded in Traditional Chinese Medicine. A new dataset consisting of 581 heart disease and 581 healthy samples was experimented by the proposed method. In order to optimize the Probabilistic Collaborative Representation Based Classifier, an analysis of its parameters was performed. According to the experimental results, the proposed method obtains the highest accuracy compared with other classifiers and is proven to be effective at heart disease detection.
BibTeX
@inproceedings{shu2017effective,
  author = {Ting Shu and Bob Zhang and Yuan Yan Tang},
  title = {{Effective Heart Disease Detection Based on Quantitative Computerized Traditional Chinese Medicine Using Representation Based Classifiers}},
  booktitle = {Evidence-based},
  year = {2017},
  doi = {10.1155/2017/7483639},
}
Figure