Interactive Segmentation Model for Placenta Segmentation from 3D Ultrasound images.

Area of Science:

• Medical Imaging
• Artificial Intelligence

Background:

• Placenta volume measurement using 3D ultrasound is crucial for predicting pregnancy outcomes.
• Manual placenta annotation is accurate but time-consuming and costly.
• Existing automated segmentation methods lack consistent robustness for clinical use.

Purpose of the Study:

• To evaluate state-of-the-art 3D interactive segmentation models for placenta segmentation.
• To assess a human-in-the-loop approach for 3D ultrasound placenta segmentation.
• To determine the efficiency of interactive models based on prompt quantity.

Main Methods:

• Comparison of publicly available 3D interactive segmentation models against a human-in-the-loop approach.
• Evaluation using Dice score, normalized surface Dice, averaged symmetric surface distance, and 95-percent Hausdorff distance.
• Assessment of model efficiency relative to the number of user prompts.

Main Results:

• The human-in-the-loop segmentation model achieved a Dice score of 0.95, meeting the success criteria.
• This model demonstrated effectiveness and efficiency in segmenting 3D ultrasound placenta images.
• Efficiency was analyzed as a function of the amount of prompts provided.

Conclusions:

• The human-in-the-loop approach provides an effective and efficient solution for interactive placenta segmentation.
• This method addresses the challenges of noise in 3D ultrasound imaging.
• The developed model offers a practical alternative to manual annotation for clinical applications.