A Multi-Task Model for Pulmonary Nodule Segmentation and Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Accurate segmentation and classification of pulmonary nodules are crucial for lung cancer diagnosis.
• Existing deep learning models often address segmentation and classification as separate tasks, neglecting potential performance gains from task correlation.

Purpose of the Study:

• To develop a unified multi-task network (MT-Net) for simultaneous pulmonary nodule segmentation and classification.
• To leverage task correlations to improve the performance of both segmentation and classification.

Main Methods:

• Proposed a multi-task network (MT-Net) with a shared backbone and prediction distillation structure.
• The MT-Net consists of coarse segmentation, classification, and fine segmentation subnetworks.
• Utilized the LIDC-IDRI dataset for quantitative and qualitative analysis.

Main Results:

• Achieved a Dice similarity coefficient (DI) of 83.2% for pulmonary nodule segmentation.
• Obtained an accuracy (ACC) of 91.9% for benign and malignant pulmonary nodule classification.
• Demonstrated competitive performance compared to state-of-the-art methods.

Conclusions:

• A unified multi-task model can effectively improve both pulmonary nodule segmentation and classification performance.
• Leveraging the correlation between segmentation and classification tasks enhances diagnostic accuracy in lung cancer detection.