A context-based chatbot surpasses trained radiologists and generic ChatGPT in following the ACR appropriateness guidelines

Abstract

AbstractBackgroundRadiological imaging guidelines are crucial for accurate diagnosis and optimal patient care as they result in standardized procedures and thus reduce inappropriate imaging studies. In the present study, we investigated the potential to support clinical decision-making using an interactive chatbot designed to provide personalized imaging recommendations based on indexed and vectorized American College of Radiology (ACR) appropriateness criteria documents.MethodsWe utilized 209 ACR appropriateness criteria documents as specialized knowledge base and employed LlamaIndex and the ChatGPT 3.5-Turbo to create an appropriateness criteria contexted chatbot (accGPT). Fifty clinical case files were used to compare the accGPT’s performance against radiologists at varying experience levels and to generic ChatGPT 3.5 and 4.0.ResultsAll chatbots reached at least human performance level. For the 50 case files, the accGPT provided a median of 83% (95% CI 82-84) ‘usually appropriate’ recommendations, while radiologists provided a median of 66% (95% CI 62-70). GPT 3.5-Turbo 70% (95% CI 67-73) and GPT 4 79% (95% CI 76-81) correct answers. Consistency was highest for the accGPT with almost perfect Fleiss’ Kappa of 0.82. Further, the chatbots provided substantial time and cost savings, with an average decision time of 5 minutes and a cost of 0.19 Euro for all cases, compared to 50 minutes and 29.99 Euro for radiologists (both p < 0.01).ConclusionChatGPT-based algorithms have the potential to substantially improve the decision-making for clinical imaging studies in accordance with ACR guidelines. Specifically, a context-based algorithm performed superior to its generic counterpart, demonstrating the value of tailoring AI solutions to specific healthcare applications.