Medical image classification by incorporating clinical variables and learned features
Jiahui Liu1, Xiaohao Cai1, Mahesan Niranjan1
1School of Electronics and Computer Science, University of Southampton, Southampton, UK.
Royal Society Open Science
|March 13, 2025
Summary
This study introduces a new method for medical image classification that combines deep learning with clinical data. The approach improves diagnostic accuracy, especially when medical data is limited.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Medical image classification is crucial for diagnostics.
- Current deep learning models often overlook valuable clinical variables.
- Limited medical data presents a significant challenge for model training.
Purpose of the Study:
- To enhance deep learning models for medical image classification by integrating clinical variables.
- To develop a method that effectively handles limited medical data.
- To provide a more comprehensive approach than single-pixel analysis.
Main Methods:
- Utilizing a pre-trained deep neural network as a feature extractor.
- Applying discriminant analysis for feature dimensionality reduction, balancing image and clinical data.
- Developing class activation maps for visualizing model focus in low-dimensional space.
Main Results:
- Demonstrated improved classification performance over state-of-the-art methods.
- Showcased effectiveness in tuberculosis and dermatology classification tasks.
- Outperformed principal component analysis in comparative evaluations.
Conclusions:
- The proposed method effectively integrates clinical variables with deep learning features for enhanced medical image classification.
- The approach addresses the challenge of data scarcity while improving diagnostic accuracy.
- Class activation maps offer valuable insights into the model's decision-making process.
