Page 109 - AIH-1-4

Artificial Intelligence in Health ChatGPT in visceral leishmaniasis diagnosis

A notable strength of this study is the use of authentic regularly audited to identify any potential bias. It is also
clinical scenarios created by an infectious disease specialist critical to ensure the transparency and accountability of
with extensive experience in the diagnosis of VL. This AI-driven diagnoses. Clinicians and patients must be able
43
approach ensures that the cases presented to ChatGPT/ to understand how the AI arrived at its conclusions to trust
GPT-4 closely resemble real-world clinical scenarios. In and effectively use these tools. It is critical to develop AI
addition, randomizing the order of case presentations and systems that provide clear and interpretable reasoning, as
using a new chat session before entering each case helped this is essential for informed decision-making.
to minimize potential biases in the AI’s responses. Another crucial issue in the field of AI-assisted medical
The potential for AI-assisted medical diagnosis to diagnosis is privacy and security, as it involves the processing
transform healthcare delivery is significant. 37,38 LLMs of sensitive patient information. 41,44,45 Robust measures
are capable of processing vast amounts of medical data must be implemented to protect data from breaches and
with remarkable speed and precision, offering several misuse. There is also a clear need to establish transparent
46
advantages over traditional diagnostic methods. 15,34,35 policies and regulations for the use and sharing of data in
46
One of the most significant advantages of AI in medical AI applications. Moreover, the potential for AI to replace
diagnosis is that it can provide diagnostic support in human clinicians raises ethical questions about the future
47
resource-limited settings where access to specialist medical of the medical profession. It must be highlighted that AI
knowledge is scarce. 21,22 AI can serve as a bridge to provide should be used to enhance and reinforce clinical decision-
expert diagnostic suggestions, thereby improving patient making, instead of replacing the critical thinking, empathy,
outcomes and healthcare efficiency. 16,25,28,37,39 Furthermore, and nuanced understanding that human clinicians can
AI-based diagnostic tools can facilitate clinicians’ provide. The role of AI in healthcare should be to assist
decision-making processes. By generating comprehensive and enhance the skills of healthcare professionals, ensuring
differential diagnosis lists, AI helps clinicians consider that patient care remains human-centered. 47
a wider range of potential conditions, thereby reducing Finally, future research should focus on expanding
the likelihood of misdiagnosis. 18,21,22,38 This is particularly the sample size and diversity of clinical cases to better
important in cases where multiple conditions may present understand the generalizability of ChatGPT’s diagnostic
with similar symptoms. capabilities. In addition, it would be beneficial to explore
Despite the encouraging results, it is important to note the integration of AI-generated diagnoses into clinical
that our study is subject to certain limitations. First, the workflows and assess the impact on clinical decision-
study employed a vignette-based methodology, 21,40 rather making and patient outcomes. Moreover, further studies
than involving real patient interactions, which may limit should also consider the potential biases and ethical
the generalizability of the findings. Second, the sample implications of using AI in healthcare. It is of the utmost
size was relatively limited, consisting of only eight clinical importance that these tools are used responsibly and
cases. The limited sample size of eight clinical cases limits equitably.
the generalizability of the study’s findings. To validate
these findings, further studies with larger and more diverse 5. Conclusion
samples are required to ensure the robustness of the This exploratory study demonstrates that ChatGPT/
conclusions. In addition, the selection of clinical vignettes GPT-4 can generate an accurate differential diagnosis for
reflecting common symptoms of VL may have contributed VL, correctly identifying the disease in a considerable
to an overestimation of the diagnostic capabilities of proportion of cases. Further research is necessary to
ChatGPT. Future studies should include a broader range confirm these findings. This study also substantiates that
of case presentations to evaluate the AI’s performance ChatGPT/GPT-4 is a promising AI-assisted diagnostic tool
in more varied clinical scenarios. Moreover, the binary with the potential to improve clinical decision-making and
scoring system used in this study, while simple, may not healthcare delivery.
fully capture the nuances of differential diagnosis accuracy.
Acknowledgments
While AI in medical diagnosis offers significant
benefits, several ethical issues must be addressed to ensure None.
its responsible use. A primary concern is the potential for
AI algorithms to reflect existing biases in medical practice Funding
and societal inequalities, which could lead to unequal This study received financial support from the Conselho
treatment. 41,42 To prevent this, it is essential that the datasets Nacional de Desenvolvimento Científico e Tecnológico
used to train AI are representative and that algorithms are (CNPq) under grant number 408003/2023-5 and from the

Volume 1 Issue 4 (2024) 103 doi: 10.36922/aih.3930