TKR-FSOD: Fetal Anatomical Structure Few-Shot Detection Utilizing Topological Knowledge Reasoning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Fetal Ultrasound Analysis

Background:

• Accurate fetal anatomical structure detection in ultrasound (US) images is crucial for diagnosis.
• Deep learning excels in US image analysis but struggles with limited data for rare conditions.
• Few-shot learning addresses data scarcity, but object detection remains underexplored in medical imaging.

Purpose of the Study:

• To propose a novel few-shot object detection method for fetal anatomical structures in ultrasound images.
• To address the challenge of detecting rare fetal conditions with limited sample data.
• To improve the accuracy and robustness of fetal ultrasound analysis.

Main Methods:

• Developed TKR-FSOD, a few-shot object detection framework for fetal ultrasound.
• Incorporated a Topological Knowledge Reasoning Module to leverage structural relationships.
• Introduced a Discriminate Ability Enhanced Feature Learning Module for robust feature extraction.

Main Results:

• TKR-FSOD significantly outperforms state-of-the-art baseline methods.
• Achieved a maximum margin of 4.8% over the second-best method on 5-shot detection.
• Demonstrated superior performance in few-shot learning scenarios for fetal anatomical structures.

Conclusions:

• The proposed TKR-FSOD method effectively addresses the challenge of few-shot object detection in fetal ultrasound.
• Leveraging topological knowledge and enhanced feature learning improves detection accuracy for limited-data categories.
• This work advances few-shot learning applications in medical image analysis, particularly for fetal anomaly detection.