A propagation-DNN: Deep combination learning of multi-level features for MR prostate segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Prostate segmentation in Magnetic Resonance (MR) imaging is challenging due to disease-induced anatomical changes and difficulties in distinguishing the prostate from adjacent tissues.
• Accurate segmentation is crucial for diagnosis and treatment planning.

Purpose of the Study:

• To develop an automated deep neural network model for precise prostate segmentation on MR images.
• To address the limitations of current segmentation techniques by integrating multi-level feature extraction.

Main Methods:

• The Propagation Deep Neural Network (P-DNN) model was developed, combining multi-level feature extraction within a single deep neural network architecture.
• High-level features were used for prostate localization and shape recognition, while low-level cues were embedded for boundary delineation via labeling propagation.

Main Results:

• The P-DNN model demonstrated superior performance compared to baseline deep neural network (DNN) models, achieving an average Dice Similarity Coefficient (DSC) improvement of 3.19%.
• On training sets, P-DNN achieved 89.9 ± 2.8% DSC and 6.84 ± 2.5 mm HD; on testing sets, it achieved 84.13 ± 5.18% DSC and 9.74 ± 4.21 mm HD, outperforming state-of-the-art non-DNN methods.

Conclusions:

• The P-DNN model effectively maximizes multi-level feature extraction for enhanced prostate segmentation in MR images.
• This automated approach offers a promising solution for improving the accuracy and efficiency of prostate MR image analysis.