Page 108 - AIH-1-4
P. 108

Artificial Intelligence in Health                                 ChatGPT in visceral leishmaniasis diagnosis




            Table 2. List of the five differential diagnoses generated by ChatGPT‑4 on March 31, 2024 for the visceral leishmaniasis clinical
            vignettes
            Case                                            Diagnoses
                         1 st              2 nd                3 rd               4 th              5 th
            01    Visceral leishmaniasis  Opportunistic infections  Systemic lupus erythematosus  Tuberculosis  Gastrointestinal infection
            02    HIV infection     Tuberculosis       Chagas disease       Visceral leishmaniasis  Leukemia or lymphoma
            03    Opportunistic infections  Lymphoma   Tuberculosis         Viral hepatitis  Chagas disease
            04    Endocarditis      Chronic liver disease  Tuberculosis     Lymphoma        HIV infection
            05    Visceral leishmaniasis  Tuberculosis  Lymphoma            Chagas disease  Typhoid fever
            06    Visceral leishmaniasis  Chagas disease  Schistosomiasis   Tuberculosis    Lymphoma
            07    Tuberculosis      Visceral leishmaniasis  Viral hepatitis  Chagas disease  Lymphoma
            08    Visceral leishmaniasis  Viral hepatitis  Chagas disease   Cirrhosis       Lymphoma
            Abbreviation: HIV: Human immunodeficiency virus.


                         A                                     B






















            Figure 1. Rates of correct diagnoses of viral leishmaniasis (VL) generated by ChatGPT/GPT-4 for each clinical vignette. (A) Shows the generated results as
            Yes or No for the diagnosis of VL in the list of five differential diagnoses. (B) Shows the position of the generated diagnosis of VL between the first and fifth
              In a different context, Biswas  explored the potential   also demonstrates their significant potential in specialized
                                      30
            of ChatGPT in radiology and found that the AI could help   medical fields, including neurology,  infectious diseases,
                                                                                           28
                                                                                                            27
            generate accurate and consistent radiology reports. This   and  clinical  scenarios.   Our  findings  and  the  literature
                                                                                 26
            corroborates our findings that AI chatbots can provide   suggest that AI chatbots have the potential to expand
            valuable diagnostic assistance in various medical fields.   beyond the diagnosis of common medical conditions and
            In addition, Cheng et al.  explored the potential of chat-  into more niche areas of medicine. This suggests that they
                                29
            GPT/GPT-4 in sports medicine, highlighting its role in   could be useful in  a wide range of healthcare settings,
            diagnostic imaging and exercise prescription. This further   thereby expanding the potential utility of AI chatbots
            underscores the broad applicability of AI in medical   beyond general practice. 15,24,25  They could also assist
            diagnostics.                                       healthcare professionals in  effectively diagnosing  a wide
              The present study focuses on the diagnosis of VL, a   range of diseases. 15,24-26
            tropical neglected disease that presents unique diagnostic   The results of this study demonstrate the potential of
            challenges. The high accuracy rates observed indicate that   ChatGPT/GPT-4 as a valuable AI-assisted diagnostic tool
            AI chatbots, such as ChatGPT/GPT-4, are not only capable   for the diagnosis of VL. By providing accurate differential
            of handling common medical conditions but also excel   diagnoses, ChatGPT/GPT-4 can assist physicians,
            in the diagnosis of more specialized and less common   especially in resource-limited settings where expert
            diseases. Other research conducted on AI-based chatbots   opinion and laboratory testing may not be readily available.


            Volume 1 Issue 4 (2024)                        102                               doi: 10.36922/aih.3930
   103   104   105   106   107   108   109   110   111   112   113