A SwinTransformer-Based Segmentation Framework With Self-Supervised Strategy for Post-Operative Prostate Cancer...

Area of Science:

• Medical Imaging
• Radiation Oncology
• Artificial Intelligence

Background:

• Radical prostatectomy for prostate cancer often necessitates postoperative radiotherapy.
• Accurate delineation of clinical target volume (CTV) and lymph node drainage area (LNA) on CT images is critical for effective radiotherapy.
• Manual segmentation of CTV and LNA is challenging and time-consuming due to anatomical changes post-surgery and lack of clear boundaries in CT images.

Purpose of the Study:

• To develop an automated and accurate method for segmenting CTV and LNA in the context of postoperative radiotherapy.
• To leverage the relationship between CTV, LNA, and organs at risk (OARs) to improve segmentation accuracy.
• To create a more efficient workflow for radiation oncologists in postoperative radiotherapy planning.

Main Methods:

• A novel cascade segmentation block was designed to integrate OARs features for guiding CTV and LNA segmentation.
• A pure SwinTransformer-based segmentation network utilizing self-supervised learning strategies was proposed.
• The model explicitly establishes correlations between CTV, LNA, and OARs.

Main Results:

• The proposed SwinTransformer-based model demonstrated superior performance compared to existing segmentation models.
• Quantitative evaluations showed higher Dice scores with reduced standard deviations.
• Qualitative assessments indicated that the segmentation results were more consistent with ground truth.

Conclusions:

• The developed AI model offers a feasible and efficient solution for CTV and LNA segmentation in postoperative radiotherapy.
• This approach can significantly streamline the radiotherapy planning process, improving efficiency for clinicians.
• The method holds promise for enhancing the precision and effectiveness of prostate cancer treatment.