An Experience-driven Interpretable Multi-task Model for Segmentation and Classification of Small Cell Lung Cancer and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading cause of cancer mortality.
• Distinguishing between small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC) is crucial for treatment and prognosis.
• Current diagnostic methods require accurate subtype classification.

Purpose of the Study:

• To develop an interpretable multi-task AI network for simultaneous segmentation and classification of SCLC and NSCLC.
• To enhance diagnostic accuracy and provide clinical insights for radiologists.

Main Methods:

• An experience-driven interpretable multi-task network based on StarNet architecture.
• Incorporation of auxiliary branches for edge uncertainty estimation and tumor core area reconstruction, leveraging SCLC characteristics.
• Utilized bottleneck layer features with a three-level contrastive loss for improved feature differentiation and interpretability.
• Implemented a feature query matching strategy for clinical insights and reference images.

Main Results:

• The proposed multi-task model outperformed single-task models in lung cancer classification and segmentation.
• The model demonstrated a degree of interpretability, aiding radiologists' decision-making.
• Experimental results validated the model's effectiveness on a public dataset.

Conclusions:

• The developed AI network offers an effective and interpretable solution for classifying SCLC and NSCLC.
• This approach has the potential to improve diagnostic accuracy and support clinical decision-making in lung cancer management.