Utilizing Large Language Models to Simplify Radiology Reports: a comparative analysis of ChatGPT3.5, ChatGPT4.0, Google Bard, and Microsoft Bing

Abstract

This paper investigates the application of Large Language Models (LLMs), specifically OpenAI's ChatGPT-3.5, ChatGPT-4.0, Google Bard, and Microsoft Bing, in simplifying radiology reports, thus potentially enhancing patient understanding. We examined 254 anonymized radiology reports from diverse examination types and used three different prompts to guide the LLMs' simplification processes. The resulting simplified reports were evaluated using four established readability indices. All LLMs significantly simplified the reports, but performance varied based on the prompt used and the specific model. The ChatGPT models performed best when additional context was provided (i.e., specifying user as a patient or requesting simplification at the 7th grade level). Our findings suggest that LLMs can effectively simplify radiology reports, although improvements are needed to ensure accurate clinical representation and optimal readability. These models have the potential to improve patient health literacy, patient-provider communication, and ultimately, health outcomes.

Comparison of large language models in management advice for melanoma: Google's AI BARD, BingAI and ChatGPT

“…examined the readability of various LLMs in radiology reports, noting how varied training data, preprocessing techniques and inherent data structures may influence each model's ability to handle unique terminologies and abbreviations. Further research could reveal the underlying differences in their code, allowing us to refine LLMs' algorithms and improve readability and comprehensibility 29 …”

“…Further research could reveal the underlying differences in their code, allowing us to refine LLMs' algorithms and improve readability and comprehensibility. 29 It is crucial to recognize that none of the models exhibited the capability to evaluate the benefits and risks associated with recommended medical procedures. This limitation becomes evident in the case of the SLNB query, where the models fail to incorporate a comprehensive discussion encompassing factors such as the individual's overall health status, personal preferences and tolerance for risk.…”

Gunning Fog Index and Automated UML Generation

“…Studies by Feng, Doshi, and Alkaldi focus on simplifying complex texts while preserving meaning, highlighting the importance of readability in text generation [24][25][26].…”

ChatGPT may free time needed by the interventional radiologist for administration / documentation: A study on the RSNA PICC line reporting template

“…As other subfields of AI, natural language processing in combination with reinforcement learning, has recently seen some remarkable advances [3], [11]. This is a broader development, not limited to the tool applied in the present study [12]. Regarding the application in radiology and elsewhere, the strengths of properly trained language processing lie in the huge knowledge base that is made available [13], and in the ability to communicate in different styles of language [14].…”