Automated Pulmonary Nodule Classification and Detection Using Deep Learning Architectures.

Area of Science:

• Biomedical imaging
• Artificial intelligence in healthcare
• Medical diagnostics

Background:

• Biomedical imaging advancements offer opportunities but analyzing large datasets is challenging for medical experts.
• Early diagnosis of lung cancer is crucial due to often asymptomatic initial stages.
• Computer-Aided Detection (CAD) systems can assist medical professionals in lung nodule classification and detection.

Purpose of the Study:

• To analyze lung cancer diagnosis through classification and detection of pulmonary nodules (benign and malignant) in CT images.
• To introduce an automated deep learning system for lung nodule classification and detection.
• To evaluate the performance of state-of-the-art deep learning detection architectures.

Main Methods:

• Utilized deep learning models including Faster-RCNN, YOLOv3, and SSD for nodule detection.
• Employed an automated deep learning system for classifying and detecting lung nodules.
• Evaluated all models on the publicly available LIDC-IDRI dataset.

Main Results:

• The automated deep learning system demonstrated enhanced accuracy in lung nodule classification and detection.
• Experimental outcomes showed a reduction in the False Positive Rate (FPR).
• State-of-the-art detection architectures were successfully applied to the LIDC-IDRI dataset.

Conclusions:

• The developed automated deep learning system effectively aids in the early diagnosis of lung cancer by classifying and detecting pulmonary nodules.
• The system shows potential in improving diagnostic accuracy and reducing false positives, supporting healthcare professionals.
• Deep learning approaches, specifically Faster-RCNN, YOLOv3, and SSD, are promising for computer-aided detection in lung cancer screening.