Artificial intelligence solution to classify pulmonary nodules on CT.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Computational pathology

Background:

• Pulmonary nodules require accurate detection and classification for timely diagnosis and treatment.
• Existing methods for nodule analysis can be time-consuming and may lack precision.

Purpose of the Study:

• To develop and evaluate a machine learning algorithm for detecting and classifying pulmonary nodules.
• To categorize nodules based on volume (>100 mm³).
• To leverage deep learning techniques for enhanced diagnostic capabilities.

Main Methods:

• A 3-stage pipeline was created: data pre-processing, nodule detection, and classification.
• Lung segmentation utilized the 3D U-NET algorithm.
• Nodule detection employed 3D Retina-UNET, with classification by a support vector machine algorithm.
• Performance was assessed using the area under the receiver operating characteristics curve (AUROC).

Main Results:

• The pipeline demonstrated strong performance in pathological nodule detection and patient diagnosis.
• An AUROC of 0.9058 was achieved for the nodule detection stage.
• The system yielded 87% accuracy, 86% specificity, and 89% sensitivity.

Conclusions:

• A functional pipeline integrating 3D U-NET, 3D Retina-UNET, and a support vector machine was successfully developed.
• The pipeline exhibits high capability for accurate pulmonary nodule classification.
• This approach offers a promising tool for improving the analysis of lung nodules.