Attention-guided deep neural network with a multichannel architecture for lung nodule classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate detection of malignant lung nodules in CT scans is critical for timely lung cancer treatment.
• Current methods may face challenges in precision and efficiency.

Purpose of the Study:

• To develop and evaluate a deep learning model with a multi-channel attention mechanism for precise malignant lung nodule diagnosis.
• To enhance the accuracy and efficiency of lung nodule classification.

Main Methods:

• Standardized CT image voxel size and generated multi-scale, multi-angle RGB images.
• Employed three attention submodels for class-specific feature extraction.
• Consolidated nodule features for final prediction, enabling dynamic localization without prior segmentation.

Main Results:

• Achieved 90.11% classification accuracy and 95.66% AUC on the LIDC-IDRI dataset.
• The model utilized only 29.09% of the time required by mainstream models.
• Demonstrated enhanced accuracy and efficiency in lung nodule classification.

Conclusions:

• The proposed attention-based deep learning model offers a significant advancement in malignant lung nodule detection.
• This approach improves diagnostic accuracy and operational efficiency for lung cancer screening.
• The model's ability to locate nodules dynamically enhances its clinical utility.