I’m trying to implement RAG for proofreading radiology reports. For now my results shows LLMs are good at tackling grammar and highlighting inconsistency between findings and impression sections. For example if “3 mm nodule” is mentioned in the findings and “30 mm nodule” in Impression section, LLama 3 was able to pick and also correct it.
However, “Contradictory” statements like “presence of Cardiomegaly. Heart size is normal” was something that it missed completely even after prompting it to pick contradictions.
Would you mind to discuss this further if you find this problem interesting?
Hi Anjali,
Thanks for posting this blog.
It’s very informative.🙌🏼
I’m trying to implement RAG for proofreading radiology reports. For now my results shows LLMs are good at tackling grammar and highlighting inconsistency between findings and impression sections. For example if “3 mm nodule” is mentioned in the findings and “30 mm nodule” in Impression section, LLama 3 was able to pick and also correct it.
However, “Contradictory” statements like “presence of Cardiomegaly. Heart size is normal” was something that it missed completely even after prompting it to pick contradictions.
Would you mind to discuss this further if you find this problem interesting?
Thanks
Param
Image Core Labs
Bangalore