Multitask connected U-Net: automatic lung cancer segmentation from CT images using PET knowledge guidance.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading cause of cancer mortality globally.
• Accurate lung tumor segmentation from medical images is crucial for diagnosis and treatment planning.
• Tumor morphology complexity presents significant segmentation challenges.

Purpose of the Study:

• To enhance lung tumor segmentation accuracy using a novel multitask connected U-Net model.
• To leverage complementary information from CT and PET imaging modalities.
• To improve segmentation performance through an integrated tumor area detection method.

Main Methods:

• Development of a multitask connected U-Net model incorporating a teacher-student framework.
• Integration of Positron Emission Tomography (PET) knowledge with Computed Tomography (CT) data.
• Implementation of a specific tumor area detection algorithm.

Main Results:

• The proposed model achieved an average Dice coefficient of 0.56 on four datasets.
• This performance surpassed existing methods including Segformer (0.51), Transformer (0.50), and UctransNet (0.43).
• Experimental results demonstrate superior lung tumor segmentation capabilities.

Conclusions:

• The proposed multitask U-Net with PET integration and tumor area detection effectively enhances lung tumor segmentation.
• This method offers a promising advancement for improving diagnostic accuracy in lung cancer care.
• The findings highlight the potential of multimodal data fusion and advanced AI models in medical image analysis.