Using Advanced Convolutional Neural Network Approaches to Reveal Patient Age, Gender, and Weight Based on Tongue
Xiaoyan Li1,2, Li Li1, Jing Wei1
1Hangzhou Normal University Affiliated Hospital, Hangzhou, Zhejiang, China.
Biomed Research International
|August 9, 2024
Summary
This study used deep convolutional neural networks (CNNs) to analyze tongue images, successfully predicting patient age and gender. Tongue analysis offers a novel, noninvasive method for health monitoring.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Biology
Background:
- The human tongue is recognized in medicine as a potential indicator of overall health.
- Traditional diagnostic methods can be invasive, costly, or inconvenient for patients.
Purpose of the Study:
- To investigate the feasibility of using deep learning models to infer patient age, gender, and weight from tongue images.
- To develop a noninvasive, accurate, and cost-effective method for patient health assessment.
Main Methods:
- Utilized pretrained deep convolutional neural networks (CNNs), specifically ResNeXt, for image analysis.
- Trained CNN models on a large dataset of dorsal tongue images.
- Evaluated model performance using metrics such as Pearson correlation coefficient, mean absolute error (MAE), accuracy, and area under the receiver operating characteristic curve (AUC).
Main Results:
- Achieved excellent age prediction accuracy (Pearson correlation coefficient = 0.71, MAE = 8.5 years).
- Demonstrated high accuracy in gender classification (80% mean accuracy, 88% AUC).
- Showed moderate accuracy for weight prediction (Pearson correlation coefficient = 0.39, MAE = 9.06 kg).
Conclusions:
- The human tongue contains significant, quantifiable information relevant to patient demographics and potentially health status.
- Deep CNNs provide a powerful tool for extracting this information from tongue images.
- This approach shows promise for noninvasive, convenient, and inexpensive patient health monitoring and diagnosis.
