Encoder-decoder with dense dilated spatial pyramid pooling for prostate MR images segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Prostate magnetic resonance (MR) image segmentation is crucial for diagnosing and treating prostate diseases.
• Challenges include low tissue contrast and small prostate regions in MR images.

Purpose of the Study:

• To develop a novel deep learning-based method for accurate and robust automatic prostate MR image segmentation.
• To address the challenges of low contrast and small effective areas in prostate MR images.

Main Methods:

• Proposed an end-to-end Encoder-Decoder network incorporating a dense dilated spatial pyramid pooling (DDSPP) module.
• The DDSPP module extracts multi-scale features, while the decoder refines prostate boundary detection.

Main Results:

• Achieved competitive results on 130 MR images, outperforming state-of-the-art methods.
• Key metrics: Dice Similarity Coefficient (DSC) of 0.954 and Hausdorff Distance (HD) of 1.752 mm.

Conclusions:

• The proposed deep learning method demonstrates high accuracy and robustness for prostate MR image segmentation.
• This approach holds significant potential for clinical applications in prostate disease diagnosis and management.