CNN-O-ELMNet: Optimized Lightweight and Generalized Model for Lung Disease Classification and Severity Assessment

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Lung diseases pose a significant global health burden, necessitating improved diagnostic tools.
• Current computer-aided detection (CAD) systems often focus on single diseases and utilize computationally intensive deep learning models.
• Existing methods struggle with efficiency and broad applicability in lung disease diagnosis.

Purpose of the Study:

• To introduce CNN-O-ELMNet, a novel, lightweight classification model for efficient and versatile lung disease detection.
• To overcome the limitations of disease-specific CAD systems and complex deep learning approaches.
• To enable early and accurate diagnosis of various lung conditions using artificial intelligence.

Main Methods:

• Developed CNN-O-ELMNet, integrating a convolutional neural network (CNN) with an optimized extreme learning machine (ELM).
• Employed the imperialistic competitive algorithm (ICA) for optimizing the extreme learning machine component.
• Evaluated the model on benchmark datasets for pneumothorax, tuberculosis, and lung cancer detection, and for multi-class severity assessment (Brixia scores).

Main Results:

• CNN-O-ELMNet achieved high accuracies in binary classifications: 97.85% for tuberculosis and 97.7% for lung cancer detection.
• The model demonstrated superior performance (p < 0.05) compared to state-of-the-art methods in specific disease classifications.
• Achieved 96.2% accuracy in multi-class assessment of lung disease severity (mild, moderate, severe) based on Brixia scores.
• Maintained low computational complexity with only 2481 trainable parameters.

Conclusions:

• CNN-O-ELMNet offers a computationally efficient and highly accurate solution for detecting multiple lung diseases.
• The model's effectiveness in binary and multi-class lung disease assessment suggests its suitability for deployment on resource-constrained healthcare devices.
• This lightweight AI approach has the potential to enhance early diagnosis and management of lung conditions globally.