Evaluation of large language models performance against humans for summarizing MRI knee radiology reports: A feasibility study