Polar transform network for prostate ultrasound segmentation with uncertainty estimation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Prostate ultrasound segmentation is challenging due to noise and unclear boundaries.
• Accurate segmentation is crucial for diagnosis and treatment planning.

Purpose of the Study:

• To develop a novel polar transform network (PTN) for accurate prostate ultrasound segmentation.
• To improve segmentation accuracy and reduce computational complexity compared to existing methods.
• To provide quantitative uncertainty estimation for clinical decision support.

Main Methods:

• Prostate segmentation in polar coordinate space using a novel Polar Transform Network (PTN).
• Parameterization of prostate surface using a 2D surface radius map.
• Incorporation of centroid perturbed test-time augmentation (CPTTA) for enhanced accuracy and uncertainty assessment.

Main Results:

• The PTN achieved superior accuracy in prostate segmentation, particularly in challenging regions like the apex and base.
• The PTN demonstrated significantly fewer trainable parameters (18%–41%) compared to convolutional neural network counterparts.
• PTN with CPTTA outperformed state-of-the-art methods statistically, with a smaller model size and reliable uncertainty estimation.

Conclusions:

• The proposed PTN offers a novel and effective approach for accurate prostate ultrasound segmentation.
• The method enhances clinical utility by providing uncertainty feedback to clinicians.
• PTN represents a significant advancement in automated medical image segmentation.