Weakly Supervised Liver Tumor Segmentation Using Couinaud Segment Annotation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate liver tumor segmentation is crucial for cancer diagnosis and treatment planning.
• Deep learning models require extensive pixel-level annotations, which are labor-intensive and limit performance.
• Radiologists commonly use Couinaud segments for tumor localization in reports.

Purpose of the Study:

• To develop a novel approach for liver tumor segmentation using less demanding Couinaud segment annotations.
• To train convolutional networks for accurate segmentation with image-level labels.

Main Methods:

• Proposed CouinaudNet model utilizes Couinaud segment annotations to generate pseudo tumor masks for supervision.
• Employs an inpainting network to remove tumors and a difference spotting network for segmentation.
• Tumor synthesis strategy generates healthy-pathological image pairs for training.

Main Results:

• The method was extensively evaluated on two liver tumor segmentation datasets.
• Achieved competitive performance compared to fully supervised methods and state-of-the-art approaches.
• Demonstrated significantly reduced annotation effort.

Conclusions:

• CouinaudNet offers an effective solution for liver tumor segmentation with reduced annotation burden.
• The approach holds promise for improving efficiency in liver cancer diagnosis and treatment planning.