AI-driven deep convolutional neural networks for chest X-ray pathology identification

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Chest X-rays are crucial for diagnosing lung diseases, but interpretation can be challenging due to inter-reader variability.
• Automated computer-aided detection systems are needed to improve accuracy and efficiency in diagnosing chest diseases from X-rays.

Purpose of the Study:

• To develop and compare the performance of deep learning models, specifically Convolutional Neural Networks (CNNs), for classifying 14 different chest diseases or pathology patterns.
• To evaluate the feasibility of transfer learning and data augmentation techniques in improving CNN performance for chest X-ray classification.

Main Methods:

• Transfer learning was employed using pre-trained CNN models (e.g., ResNet-152, Inception-V3) on a large dataset of 91,324 chest X-ray images.
• A deep convolution generative adversarial network (DC-GAN) was utilized for data augmentation to address small or imbalanced datasets by generating synthetic images.
• Eight different CNN models were trained and evaluated for their classification accuracy.

Main Results:

• ResNet-152 with data augmentation achieved the highest accuracy of 83% for classifying six disease classes.
• Without data augmentation, ResNet-152 achieved 67% accuracy, while other models like Inception-V3 reached 68% accuracy.
• The study demonstrated the effectiveness of GAN-based data augmentation in improving model performance.

Conclusions:

• The study successfully demonstrated the feasibility of using transfer learning CNN methods for classifying 14 types of chest diseases from X-ray images.
• GAN-based techniques effectively addressed data scarcity issues, enhancing the classification performance.
• These automated schemes can assist healthcare professionals in more accurate and efficient lung disease detection.