Hybrid transformer-CNN and LSTM model for lung disease segmentation and classification

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Pulmonary Disease Diagnostics

Background:

• Lung disorders are a leading cause of mortality globally, necessitating improved diagnostic methods.
• Accurate segmentation and classification of lung diseases from medical images are challenging due to pathological variability.
• Early and precise diagnosis is crucial for effective pulmonary disease management.

Purpose of the Study:

• To propose a novel hybrid LinkNet-Modified LSTM (L-MLSTM) model for lung disease segmentation and classification.
• To enhance the accuracy and efficiency of pulmonary disease diagnosis using deep learning techniques.
• To address the challenges posed by diverse lung pathologies in medical image analysis.

Main Methods:

• Image pre-processing using median filtering.
• Segmentation of affected lung regions using an improved Transformer-based CNN (ITCNN).
• Feature extraction including texture, shape, color, and deep features.
• Classification using the hybrid L-MLSTM model on two CT scan datasets.

Main Results:

• The L-MLSTM model achieved high accuracies of 89% and 95% on two distinct datasets.
• Demonstrated superior performance compared to existing models like HDE-NN, DBN, LSTM, and CNN.
• The hybrid approach effectively handles complexity and variability in lung images, reducing diagnostic errors.

Conclusions:

• The proposed L-MLSTM model offers a robust and accurate solution for lung disease segmentation and classification.
• This deep learning framework has the potential to significantly improve early diagnosis and treatment of pulmonary conditions.
• The hybrid model's performance highlights its capability in distinguishing various lung diseases from medical images.