Automated subcutaneous fat segmentation with a convolutional neural network in magnetic resonance guided high-intensity focused ultrasound treatment for uterine fibroids.

Key Points

• To develop an automated algorithm for accurately segmenting abdominal fat in MR images of uterine fibroid patients.
• Used an attention-gated U-Net architecture for segmentation.
• Analyzed MR images of patients with uterine fibroids.
• Evaluated the model's accuracy against data prepared by a second reader.
• The algorithm achieved high segmentation accuracy.
• It effectively handled varying thicknesses of subcutaneous fat layers.
• The model performed reliably across multiple datasets.