Fig 6From: Exploring the performance and explainability of fine-tuned BERT models for neuroradiology protocol assignmentThe bar plot decomposes the mistakes into four categories: multifarious choices, age-related, ambiguous text, and flagrant errorsBack to article page