To investigate whether a large language model (LLM) can accurately predict histopathological types of breast cancer based on imaging findings from contrast-enhanced breast MRI reports.
Key Findings:
180 cases were included, resulting in 186 lesions analyzed.
Pathological results classified into 10 histopathological diagnoses.
LLM's predictions were assessed against pathology reports for accuracy.
Interpretation:
The study explores the potential of LLMs to bridge imaging findings and histopathological diagnoses, which could enhance reporting quality and facilitate research.
Limitations:
Retrospective design may introduce bias.
Exclusion of certain cases limits generalizability.
The LLM's predictions were not based on pathology report text.
Conclusion:
The ability of LLMs to predict histopathological types from MRI findings represents a promising step towards integrating AI into radiology workflows, though further validation is needed.
