[An automatic pulmonary nodules detection algorithm with multi-scale information fusion].

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung nodules are primary indicators of early-stage lung cancer.
• Accurate and rapid detection of pulmonary nodules in CT images is challenging due to complex backgrounds and varied nodule characteristics.
• Early diagnosis and treatment of lung cancer significantly improve patient outcomes.

Purpose of the Study:

• To develop and validate a multi-scale feature fusion algorithm for automatic and accurate pulmonary nodule detection.
• To enhance the precision of identifying lung nodules in computed tomography (CT) images.

Main Methods:

• A three-layer modular deep convolutional neural network (VGG16) was designed for nodule detection.
• The model employs feature extraction, multi-scale feature fusion, and analysis of fused features to identify candidate nodule regions.
• Non-maximum suppression was utilized to refine candidate boxes and determine final nodule locations.

Main Results:

• The proposed algorithm achieved an average detection accuracy of 90.9% for pulmonary nodules.
• Validation was performed using the LIDC-IDRI dataset, a common benchmark for lung nodule analysis.
• The multi-scale approach effectively enhanced nodule details and improved detection performance.

Conclusions:

• The developed multi-scale feature fusion algorithm demonstrates high accuracy in automatic pulmonary nodule detection.
• This method offers a promising approach for improving early lung cancer diagnosis through enhanced CT image analysis.
• The algorithm's performance suggests its potential utility in clinical settings for lung cancer screening.