Impact of Annotation Level on Multisequence MRI Models for Preoperative Microvascular Invasion Prediction in Hepatocellular Carcinoma

Key Points

• The research aims to evaluate the effect of annotation levels on the prediction accuracy of preoperative microvascular invasion in hepatocellular carcinoma using MRI models.
• Comparison of bounding box versus voxel-level mask annotations.
• Application of multisequence MRI techniques for model training.
• Statistical analysis of prediction performance based on different annotation levels.
• Bounding box annotations achieve performance comparable to voxel-level masks.
• Significant improvement in annotation efficiency was observed using bounding box methods.