Ensemble based in transfer learning for cytological classification in pleural fluid.

Area of Science:

• Computational pathology
• Artificial intelligence in diagnostics
• Medical image analysis

Background:

• Manual interpretation of pleural effusion cytology is time-consuming and subjective.
• Automated solutions are needed, especially for malignant pleural effusion in resource-limited settings.

Purpose of the Study:

• To develop and evaluate an ensemble deep learning framework for classifying pleural cytology images.
• To assess the impact of data augmentation on model performance.

Main Methods:

• An ensemble model combining ResNet50V2, DenseNet121, and InceptionV3 was trained using transfer learning.
• Three data augmentation scenarios (0%, 50%, 300%) were tested on local and external datasets.
• Performance was evaluated using accuracy, precision, recall, and F1-score.

Main Results:

• The ResNet+DenseNet ensemble with 300% data augmentation achieved 96.2% accuracy on the local dataset and 89.6% on the external dataset.
• The ensemble model outperformed individual architectures.
• Increased data augmentation improved model generalization and robustness.

Conclusions:

• Ensemble deep learning models with data augmentation provide accurate and reproducible diagnostic support for pleural cytology.
• This approach has potential for deployment in low-resource settings to improve cancer diagnosis accessibility.