Pelvic bone tumor segmentation fusion algorithm based on fully convolutional neural network and conditional random...

Area of Science:

• Orthopedic oncology
• Medical image analysis
• Deep learning

Background:

• Pelvic bone tumors, both benign and malignant, pose significant orthopedic challenges.
• Current machine learning algorithms for bone tumor segmentation exhibit limited accuracy.
• Accurate segmentation is crucial for diagnosis and treatment planning.

Purpose of the Study:

• To develop an enhanced bone tumor image segmentation algorithm for improved accuracy.
• To address the limitations of existing machine learning models in segmenting pelvic bone tumors.
• To achieve precise and stable segmentation of pelvic bone tumors.

Main Methods:

• An enhanced fully convolutional neural network (FCNN-4s) was utilized for initial segmentation.
• Batch normalization layers were incorporated to accelerate convergence and improve model accuracy.
• A fully connected conditional random field (CRF) was integrated for refining segmentation boundaries.

Main Results:

• The enhanced algorithm demonstrated significant improvements in segmentation accuracy and stability.
• The model achieved an average Dice coefficient of 93.31%, indicating superior performance.
• The algorithm effectively addressed over-segmentation and under-segmentation issues.

Conclusions:

• The proposed segmentation model offers superior real-time performance and robust stability.
• The enhanced FCNN-4s with CRF integration achieves heightened segmentation accuracy for pelvic bone tumors.
• This advanced algorithm represents a significant advancement in orthopedic image analysis.