Classification of distinct lung diseases using novel enhanced long short-term memory based optimization methodology

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Chest X-ray (CXR) classification is vital for early lung disease diagnosis.
• Challenges include low image quality, similar disease features, and unstable classification.
• Existing methods require improvement for reliable computer-aided diagnosis.

Purpose of the Study:

• To propose a novel ELSTM-AZOA framework for multiclass lung disease classification using CXR images.
• To enhance the accuracy and reliability of computer-aided lung disease diagnosis.
• To classify six categories: healthy lung, tuberculosis, pneumonia, lung cancer, COPD, and COVID-19.

Main Methods:

• Preprocessing CXR images using balance contrast enhancement.
• Lung region segmentation with U-Net++.
• Feature extraction via statistical and gray level co-occurrence matrix methods.
• Classification using an enhanced long short-term memory (ELSTM) network.
• Model parameter optimization with the American zebra optimization algorithm (AZOA).

Main Results:

• The ELSTM-AZOA framework achieved superior performance compared to existing methods.
• Demonstrated a 6.36% increase in accuracy and a 6.43% increase in precision.
• Successfully classified six distinct lung conditions from CXR images.

Conclusions:

• The proposed ELSTM-AZOA framework offers a robust and reliable method for computer-aided lung disease classification.
• The findings highlight the potential of the framework for improving clinical decision-making.
• This approach shows promise for enhancing diagnostic accuracy in medical imaging analysis.