Pediatric pneumonia diagnosis using stacked ensemble learning on multi-model deep CNN architectures.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatrics

Background:

• Pediatric pneumonia presents a significant global health challenge with high mortality rates.
• Chest X-rays are crucial for diagnosis but face limitations in accuracy for children due to low radiation levels.
• Accurate and timely diagnosis of pediatric pneumonia is essential for effective treatment and improved outcomes.

Purpose of the Study:

• To develop and validate an accurate computer-aided diagnosis (CAD) model for pediatric pneumonia.
• To enhance the diagnostic capabilities of chest X-rays using advanced image processing and machine learning techniques.
• To address the challenges in detecting pediatric pneumonia in low-radiation X-ray images.

Main Methods:

• Image enhancement using Contrast Limited Adaptive Histogram Equalization (CLAHE).
• Feature extraction from deep learning models (MobileNet, DenseNet121, DenseNet169, DenseNet201).
• A stacking ensemble classifier fusing deep learning features for classification, validated with Stratified K-Fold cross-validation.

Main Results:

• The proposed model achieved high performance metrics: 98.62% accuracy, 98.99% precision, 99.53% recall, and 99.26% F1 score.
• An AUC score of 93.17% was obtained, demonstrating strong discriminative power.
• The stacking classifier effectively integrated features from multiple deep learning architectures.

Conclusions:

• The developed CAD model shows exceptional performance in diagnosing pediatric pneumonia from chest X-rays.
• This approach offers a promising tool for real-time, accurate, and reliable diagnosis of pediatric pneumonia.
• The model has the potential to significantly assist clinicians in managing pediatric pneumonia cases.