Deep Learning-Based Classification of Chest Diseases Using X-rays, CT Scans, and Cough Sound Images.

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Respiratory Medicine

Background:

• Chest diseases share similar symptoms, complicating accurate diagnosis by radiologists and health experts.
• Effective identification of diverse chest conditions like lung cancer, COVID-19, pneumonia, and tuberculosis is crucial.

Purpose of the Study:

• To develop and evaluate a novel deep learning (DL) model for identifying nine distinct chest diseases.
• To enhance the accuracy and efficiency of chest disease classification using multimodal data.

Main Methods:

• A deep learning-based chest disease detection network (DCDD_Net) was designed, utilizing Chest X-rays (CXR), CT scans, and cough sound images.
• Cough sounds were converted into images using the scalogram method, and borderline SMOTE was applied for data balancing.
• The DCDD_Net model was trained and validated on 20 public benchmark datasets, comparing its performance against established models.

Main Results:

• The DCDD_Net model achieved high performance metrics, including 96.67% accuracy, 96.82% precision, 95.76% recall, 95.61% F1-score, and 99.43% AUC.
• Statistical evaluations using McNemar and ANOVA tests confirmed the model's robustness and resilience.
• DCDD_Net significantly outperformed baseline models like InceptionResNet-V2, EfficientNet-B0, DenseNet-201, and Xception.

Conclusions:

• The DCDD_Net model demonstrates superior performance in classifying nine different types of chest diseases.
• This deep learning approach offers significant potential to assist radiologists and medical professionals in disease detection.
• The study highlights the effectiveness of integrating multimodal data (CXR, CT, cough sounds) for improved diagnostic accuracy.