Rethinking automatic segmentation of gross target volume from a decoupling perspective

Area of Science:

• Medical image analysis
• Artificial intelligence in oncology
• Radiation therapy planning

Background:

• Manual Gross Target Volume (GTV) segmentation for radiation therapy (RT) is labor-intensive and prone to inconsistencies.
• Existing deep learning models struggle with GTV segmentation due to low image contrast and severe pixel imbalance.

Purpose of the Study:

• To develop an automated GTV segmentation method that overcomes limitations of current approaches.
• To improve the accuracy and reliability of GTV delineation in cancer RT planning.

Main Methods:

• Proposed a Heterogeneous Cascade Framework (HCF) decoupling GTV segmentation into recognition and segmentation subtasks.
• Introduced a multi-level Spatial Alignment Network (SANet) with a spatial alignment module to mitigate information loss.
• Implemented a Combined Regularization (CR) loss and Balance-Sampling Strategy (BSS) to address pixel imbalance and enhance convergence.

Main Results:

• The HCF demonstrated superior performance compared to state-of-the-art methods on the StructSeg2019 challenge datasets.
• Achieved significant reductions in false positives and improved segmentation of small objects.
• The method effectively handles low contrast and pixel imbalance challenges inherent in GTV segmentation.

Conclusions:

• The proposed HCF offers a robust and efficient solution for automated GTV segmentation in RT planning.
• This framework has the potential to enhance the precision and reduce variability in cancer treatment planning.
• The method shows promise for clinical application, improving workflow efficiency and patient outcomes.