ZooCNN: A Zero-Order Optimized Convolutional Neural Network for Pneumonia Classification Using Chest Radiographs.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Pneumonia is a major cause of child mortality, especially in Least Developed Countries (LDCs).
• Current diagnosis relies on chest X-rays (CXRs), but radiologist shortages in LDCs hinder access.
• Automated diagnostic systems are crucial for improving pneumonia detection in underserved regions.

Purpose of the Study:

• To develop and evaluate a Deep Learning model for automated pneumonia classification from CXR images.
• To address challenges of conventional Convolutional Neural Networks (CNNs) like overfitting and high computational costs.
• To create an efficient and accurate tool for distinguishing Normal Lungs (NL), Bacterial Pneumonia (BP), and Viral Pneumonia (VP).

Main Methods:

• A novel Zero-Order Optimized Convolutional Neural Network (ZooCNN) was developed using Zero-Order Optimization (Zoo).
• The Adaptive Synthetic Sampling (ADASYN) approach was employed for class balancing on the Kaggle CXR Images dataset.
• Hyperparameter finetuning was performed using the ZooPlatform (ZooPT), resulting in a 72% weight reduction.

Main Results:

• The ZooCNN model achieved high performance metrics: 97.27% accuracy, 97.00% sensitivity, 98.60% specificity, and 97.03% F1 score.
• The model demonstrated superior efficacy compared to contemporary pneumonia classification models.
• The optimized architecture significantly reduced model weights, enhancing computational efficiency.

Conclusions:

• The ZooCNN presents a highly accurate and efficient automated solution for pneumonia classification from CXR images.
• This model has the potential to significantly support clinical decision-making, particularly in LDCs with limited radiologist access.
• The developed approach offers a promising tool for improving pediatric pneumonia diagnosis and patient outcomes.