A new AI-based diagnostic tool, called the Bias-Free Network (RBF-Net), offers a solution to a major challenge in cough-based diagnosis of respiratory diseases (RDs). Many existing models overlook confounding variables, such as age, gender, and smoking status, which can lead to biased predictions and unreliable performance. To tackle this issue, RBF-Net uses a hybrid of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, along with a bias predictor that employs a conditional Generative Adversarial Network (c-GAN). This setup helps mitigate the impact of confounders in the training data, ensuring more accurate RD diagnosis.
Incorporating COVID-19 data, RBF-Net was tested on unbalanced datasets with confounding factors, including gender, age, and smoking status. It achieved test set accuracies of 84.1%, 84.6%, and 80.5%, respectively, outperforming a traditional CNN-LSTM model by 5.5%, 7.7%, and 8.2%.
The results highlight RBF-Net’s robustness in biased training conditions, offering a significant improvement in reliability for AI-driven respiratory disease diagnostics.
This advancement holds promise for more accurate, unbiased predictions, providing valuable support for healthcare professionals.
To Know More: https://doi.org/10.3390/bioengineering11010055
Reference: Saeed T, et al. An AI-enabled bias-free respiratory disease diagnosis model using cough audio. Bioengineering. 2024 Jan 5;11(1):55.