An Optimal Model Combining SqueezeNet and Machine Learning Methods for Lung Disease Diagnosis

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Machine Learning for Disease Diagnosis

Background:

• Artificial intelligence (AI) is transforming healthcare, particularly medical imaging, by enhancing diagnostic accuracy and enabling patient self-diagnosis.
• Challenges remain in distinguishing similar lung conditions using chest X-rays, despite the utility of machine learning datasets.
• Integrating AI offers potential for improved healthcare outcomes and operational efficiency.

Purpose of the Study:

• To present a novel AI model capable of differentiating, diagnosing, and classifying three distinct diseases with highly similar symptoms from chest X-rays.
• To reduce the number of parameters in the AI model while maintaining high diagnostic precision for application in embedded systems.

Main Methods:

• The proposed model integrates the SqueezeNet neural network architecture with machine learning classifiers: logistic regression, support vector machine (SVM), k-nearest neighbors (KNN), decision tree, and naive Bayes.
• A dataset of chest X-ray (CXR) images was utilized, classified into four categories: pneumonia, tuberculosis, COVID-19, and normal cases.

Main Results:

• The AI model achieved high performance metrics: 97.32% accuracy, 97.33% precision, 97.31% F1 score, 97.30% recall, and 99.40% AUC.
• The model employs a significantly reduced parameter count (4.6 million) compared to existing state-of-the-art models (47 million).

Conclusions:

• The developed AI model demonstrates strong classification accuracy for differentiating lung conditions.
• The model's reduced parameter count leads to lower internal memory and computing resource requirements, making it suitable for embedded systems.