Related Experiment Video
Updated: Jan 16, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
7.4K
A Comparative Evaluation of Meta-Learning Models for Few-Shot Chest X-Ray Disease Classification
Luis-Carlos Quiñonez-Baca1, Graciela Ramirez-Alonso1, Fernando Gaxiola2
1Computer Vision and Data Science Lab, Facultad de Ingeniería, Universidad Autónoma de Chihuahua, Circuito Universitario Campus II, Chihuahua 31125, Mexico.
Diagnostics (Basel, Switzerland)
|September 27, 2025
Summary
Meta-learning effectively classifies thoracic diseases from limited chest X-ray data. Prototype-based approaches, like Prototypical Networks with DenseNet-121, offer robust and efficient few-shot learning for medical imaging diagnostics.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Limited labeled data hinders deep learning for medical diagnostics, especially for rare diseases.
- Deep learning models require large annotated datasets, which are often unavailable in healthcare.
- Meta-learning enables rapid adaptation to new tasks with minimal labeled data.
Purpose of the Study:
- To evaluate meta-learning models for thoracic disease classification using chest X-rays.
- To compare the performance of Prototypical Networks, Relation Networks, MAML, and FoMAML.
- To identify optimal backbone architectures for meta-learning in this context.
Main Methods:
- Comparative evaluation of four meta-learning algorithms: Prototypical Networks, Relation Networks, MAML, and FoMAML.
- Assessment of five backbone architectures (ConvNeXt, DenseNet-121, ResNet-50, MobileNetV2, ViT) with Prototypical Networks.
- Experiments conducted on the ChestX-ray14 dataset using a 2-way, k-shot setting.
Main Results:
- Prototypical Networks with DenseNet-121 yielded the best performance (Recall: 68.1%, F1: 67.4%, Precision: 0.693) in the 2-way, 10-shot configuration.
- Hernia classification achieved the highest accuracy in disease-specific analysis.
- Prototypical and Relation Networks showed superior computational efficiency (fewer FLOPs, shorter execution times) compared to MAML and FoMAML.
Conclusions:
- Prototype-based meta-learning, especially with DenseNet-121, is a robust and efficient method for few-shot chest X-ray classification.
- This approach shows significant potential for clinical applications with scarce annotated medical data.
- Meta-learning offers a viable solution to overcome data limitations in medical AI development.
