An efficient interactive segmentation framework for medical images without pre-training.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate medical image segmentation is crucial for clinical diagnosis and surgical planning.
• Existing methods often require extensive training data, limiting their applicability.

Purpose of the Study:

• To propose an efficient interactive framework for medical image segmentation using a graph convolutional network (GCN).
• To enable accurate segmentation with minimal user interaction and without requiring pre-existing training data.

Main Methods:

• Developed an adaptive interactive segmentation approach allowing users to guide segmentation by clicking or correcting control points.
• Introduced an interactive segmentation network (IVIF-GCN) that learns from user interactions, transforming cues into annotations.
• Proposed an information fusion module (IVIF) within IVIF-GCN to integrate image and vertex position features for improved localization.

Main Results:

• Achieved high performance with mean Dice scores of 96.6% on the PROMISE12 dataset and 91.3% on a nasopharyngeal carcinoma (NPC) dataset.
• Demonstrated superior performance compared to state-of-the-art medical image segmentation methods.

Conclusions:

• The proposed interactive segmentation method enhances clinical application results efficiently, even without training data.
• A graphical user interface (GUI) tool implementing this method is publicly available for broader use.