Establishing an AI-based diagnostic framework for pulmonary nodules in computed tomography

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Pulmonary nodules detected via computed tomography (CT) can be benign or malignant, necessitating early detection for effective management.
• Current manual methods for identifying pulmonary nodules are inefficient, prone to errors, and time-consuming.

Purpose of the Study:

• To develop an Artificial Intelligence (AI) diagnostic framework to enhance the identification and categorization of pulmonary nodules from CT scans.
• To improve diagnostic accuracy and efficiency in pulmonary nodule analysis.

Main Methods:

• A deep learning framework utilizing convolutional neural networks was employed, processing 1,056 3D-DICOM CT images.
• The framework included preprocessing steps like lung segmentation, nodule detection using Retina-UNet, and classification with a Support Vector Machine (SVM).
• Model performance was evaluated using sensitivity, specificity, and Area Under the Receiver Operating Characteristic curve (AUROC).

Main Results:

• The developed AI model achieved an AUROC of 0.9058, indicating strong performance.
• Diagnostic accuracy reached 90.58%, with a positive predictive value of 89% and a negative predictive value of 86%.
• The AI algorithm demonstrated effective image preprocessing and robust nodule detection and classification capabilities.

Conclusions:

• The AI-based diagnostic framework significantly improved diagnostic accuracy for pulmonary nodules compared to traditional methods.
• The AI system offers high reliability in detecting and classifying pulmonary nodules, reducing inter-observer variability and potentially improving clinical outcomes.
• Future work may involve expanding the annotated dataset and refining the model to address challenges with non-solitary nodule detection.