Page 105 - AIH-1-4

Artificial Intelligence in Health ChatGPT in visceral leishmaniasis diagnosis

In the context of the evolving artificial intelligence (AI) study involved the use of clinical vignette cases and thus
landscape, tools such as large language models (LLMs) have did not require individual informed consent, approval was
been applied to a range of healthcare activities, including obtained from the Ethics Committee of the HU-UNIVASF
answering medical questions, examination, and diagnosis (approval number 6.967.834) to ensure that the study was
in hospitals. 15,16 In this sense, LLMs such as Chat Generative conducted according to the highest standards of research
Pre-trained Transformer (ChatGPT) offer significant integrity.
potential to aid in medical diagnosis. 15,17,18 ChatGPT, an
AI chatbot running on a transformer architecture, 19,20 has 2.2. Case materials
shown promise in interpreting and integrating medical A Brazilian infectious disease physician, an expert in the
data to assist healthcare professionals. While not a diagnosis of VL and other neglected tropical diseases,
15
substitute for clinical judgment, AI tools can enhance the formulated eight Brazilian clinical case studies. These
diagnostic process by providing differential diagnoses and studies were based on the common sociodemographic and
refining clinical suspicions. 21,22 clinical characteristics of outpatients diagnosed with VL at
HU-UNIVASF.
In the dynamic field of AI, ChatGPT represents a
significant advancement in improving human-machine The infectious disease specialist, a third investigator
communication. 18,23 By employing deep learning principles, in this study (S.R.deA.), wrote the eight clinical vignette
ChatGPT leverages a comprehensive neural network cases in Brazilian Portuguese. Each vignette included
model that is capable of understanding and generating text information on the patient’s age, biological sex, place of
with nuanced context, tone, and intent. 19,20 The application birth, and other sociodemographic indicators identifying
of ChatGPT in medicine is growing, with research information, social history, history of present illness, past
highlighting its utility in patient education, interaction, and medical history, and physical examination.
health information dissemination. 15,18,22,24-31 In particular, It is important to emphasize that the infectious disease
ChatGPT has shown varying degrees of accuracy in physician did not change the geographical scope of the
diagnosing medical conditions, underscoring its potential study (São Francisco Valley, Brazil). Table 1 describes
role in the complex disease diagnostic processes. 17,21,22,26,32,33 the eight clinical vignette cases used in the present study.
However, despite its potential, research on the application The original texts, written by the physician in Brazilian
of ChatGPT for the diagnostics of neglected tropical Portuguese, were translated into English for presentation
diseases remains limited, especially for the diagnosis of VL. in this study.
In this sense, this exploratory study evaluates the
diagnostic accuracy of differential diagnosis lists generated 2.3. Diagnosis lists generated by ChatGPT-4
by ChatGPT for clinical vignette cases of VL. Given the On March 31, 2024, we utilized the ChatGPT (GPT 4.0,
complexity of VL diagnosis, this research explores the https://chatgpt.com/, OpenAIOpCo, LLC, San Francisco,
integration of AI tools to assist healthcare professionals in CA, USA) for our research. ChatGPT-4 is an advanced
making accurate and timely diagnoses to improve patient natural language processing chatbot developed by OpenAI
outcomes. In addition, this study aims to fill this gap by OpCo, LLC that builds on the success of previous models
exploring how ChatGPT/GPT-4 can effectively integrate such as GPT-3. ChatGPT/GPT-4 is an LLM that has been
various medical data into the assisted diagnostic process trained on large amounts of text data, enabling it to generate
of VL. human-like responses across multiple domains. 15,34,35 The
findings from the Global Burden of Disease Study have
2. Data and methods been integrated into the ChatGPT/GPT-4 platform to
enhance personalized healthcare planning through the
2.1. Study design
use of AI-assisted disease burden assessment and planning
In this exploratory study, we evaluated the diagnostic tools. 32
accuracy of differential diagnosis lists generated by The differential diagnoses generated by ChatGPT/
ChatGPT (GPT 4.0, https://chatgpt.com/, OpenAI OpCo, GPT-4 are derived from its extensive training on a wide
LLC, San Francisco, CA, United States of America [USA]) array of medical literature, including significant studies
for clinical vignette cases of VL formulated in Brazilian. such as the Global Burden of Disease study. It must
36
Portuguese on March 31, 2024, the study was conducted also be noted that in this study, ChatGPT/GPT-4 did not
at the Dr. Washington Antônio de Barros Teaching access external sources in real time; instead, it produced
34
Hospital (HU-UNIVASF) of the Brazilian Hospital Services responses solely based on the knowledge acquired during
Company in Petrolina, Pernambuco, Brazil. Although the its training phase. 15

Volume 1 Issue 4 (2024) 99 doi: 10.36922/aih.3930