Prior knowledge-based multi-task learning network for pulmonary nodule classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Pulmonary nodule classification relies on morphological attributes, but existing models often overlook attribute correlations.
• Accurate classification of benign versus malignant pulmonary nodules is critical for patient outcomes.

Purpose of the Study:

• To develop an advanced pulmonary nodule classification model that incorporates inherent attribute correlations.
• To improve the accuracy and comprehensiveness of pulmonary nodule diagnosis using a novel network architecture.

Main Methods:

• Proposed a prior knowledge-based multi-task learning (PK-MTL) network.
• Treated attribute correlations as prior knowledge, encoded using hypergraph neural networks and multi-order task transfer learning.
• Integrated segmentation, classification, and attribute scoring within a multi-task framework with feature fusion and attention blocks.

Main Results:

• Achieved 91.04% accuracy in diagnosing malignant pulmonary nodules on the LIDC-IDRI dataset.
• Demonstrated state-of-the-art performance, outperforming existing pulmonary nodule classification models.
• The PK-MTL network effectively leverages attribute correlations for improved diagnostic accuracy.

Conclusions:

• The proposed PK-MTL network offers a significant advancement in pulmonary nodule classification by integrating attribute correlations.
• This approach enhances diagnostic accuracy and provides a more comprehensive analysis for radiologists.
• The method holds promise for improving early detection and treatment planning for lung cancer.