An Effective MR-Guided CT Network Training for Segmenting Prostate in CT Images

Area of Science:

• Medical Imaging
• Radiotherapy
• Deep Learning

Background:

• Prostate segmentation in CT images is challenging due to low tissue contrast.
• Existing methods using CT alone have limited performance.
• MRI offers better soft tissue contrast for prostate segmentation.

Purpose of the Study:

• To improve prostate segmentation accuracy in CT images using indirect guidance from MRI.
• To introduce a novel deep transfer learning approach, MICS-NET, for MR-guided CT network training.

Main Methods:

• Proposed MICS-NET, a deep transfer learning approach for MR-guided CT prostate segmentation.
• Implemented a two-step guidance process: feature transfer from MRI to CT and adaptive likelihood transfer with view consistency.
• Evaluated MICS-NET on a real CT prostate image dataset with manual delineations as ground truth.

Main Results:

• MICS-NET achieved a 6% higher Dice Ratio compared to a CT-only segmentation model.
• The MR-guided approach demonstrated performance comparable to models trained solely on MRI data.
• Promising segmentation results were generated for CT prostate images.

Conclusions:

• MR-guided CT network training (MICS-NET) effectively improves prostate segmentation in CT images.
• Deep transfer learning from MRI provides valuable guidance for enhancing CT-based segmentation.
• The proposed method offers a viable solution for improving radiotherapy planning through accurate prostate segmentation.