Deep learning classifiers for computer-aided diagnosis of multiple lungs disease.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Deep Learning for Disease Diagnosis

Background:

• Computer-aided diagnosis (CAD) is rapidly advancing due to deep learning, big data, and enhanced computational power.
• CAD offers cost-effective and safer alternatives to traditional diagnostic methods.

Purpose of the Study:

• To develop an efficient CAD system for diagnosing three common diseases: lung cancer, pneumonia, and COVID-19.
• To evaluate the performance of different deep learning models for multi-disease classification using X-ray images.

Main Methods:

• Three deep learning models (Inception V3, CNN, LSTM) were designed for 4-way classification.
• Publicly available datasets for lung cancer, COVID-19, and pneumonia were utilized.
• Class imbalance was addressed using pre-processing and data augmentation, resulting in 1386 subjects per class.

Main Results:

• The combined CNN-LSTM model demonstrated superior performance with 94.5% accuracy.
• This accuracy surpassed that of individual CNN and InceptionV3-LSTM models.
• Results were validated using 3, 5, and 10-fold cross-validation.

Conclusions:

• A single CAD system can effectively diagnose multiple diseases.
• The CNN-LSTM model shows significant promise for multi-disease diagnosis in medical imaging.