Optimizing GPT-4 Turbo Diagnostic Accuracy in Neuroradiology through Prompt Engineering and Confidence Thresholds.

AI diagnostic tools GPT-4 Turbo artificial intelligence in medicine clinical decision support confidence thresholds diagnostic imaging large language model (LLM) misdiagnosis reduction neuroradiology prompt engineering

Journal

Diagnostics (Basel, Switzerland)
ISSN: 2075-4418
Titre abrégé: Diagnostics (Basel)
Pays: Switzerland
ID NLM: 101658402

Informations de publication

Date de publication:
17 Jul 2024
Historique:
received: 29 05 2024
revised: 02 07 2024
accepted: 10 07 2024
medline: 27 7 2024
pubmed: 27 7 2024
entrez: 27 7 2024
Statut: epublish

Résumé

Integrating large language models (LLMs) such as GPT-4 Turbo into diagnostic imaging faces a significant challenge, with current misdiagnosis rates ranging from 30-50%. This study evaluates how prompt engineering and confidence thresholds can improve diagnostic accuracy in neuroradiology. We analyze 751 neuroradiology cases from the American Journal of Neuroradiology using GPT-4 Turbo with customized prompts to improve diagnostic precision. Initially, GPT-4 Turbo achieved a baseline diagnostic accuracy of 55.1%. By reformatting responses to list five diagnostic candidates and applying a 90% confidence threshold, the highest precision of the diagnosis increased to 72.9%, with the candidate list providing the correct diagnosis at 85.9%, reducing the misdiagnosis rate to 14.1%. However, this threshold reduced the number of cases that responded. Strategic prompt engineering and high confidence thresholds significantly reduce misdiagnoses and improve the precision of the LLM diagnostic in neuroradiology. More research is needed to optimize these approaches for broader clinical implementation, balancing accuracy and utility.

Sections du résumé

BACKGROUND AND OBJECTIVES OBJECTIVE
Integrating large language models (LLMs) such as GPT-4 Turbo into diagnostic imaging faces a significant challenge, with current misdiagnosis rates ranging from 30-50%. This study evaluates how prompt engineering and confidence thresholds can improve diagnostic accuracy in neuroradiology.
METHODS METHODS
We analyze 751 neuroradiology cases from the American Journal of Neuroradiology using GPT-4 Turbo with customized prompts to improve diagnostic precision.
RESULTS RESULTS
Initially, GPT-4 Turbo achieved a baseline diagnostic accuracy of 55.1%. By reformatting responses to list five diagnostic candidates and applying a 90% confidence threshold, the highest precision of the diagnosis increased to 72.9%, with the candidate list providing the correct diagnosis at 85.9%, reducing the misdiagnosis rate to 14.1%. However, this threshold reduced the number of cases that responded.
CONCLUSIONS CONCLUSIONS
Strategic prompt engineering and high confidence thresholds significantly reduce misdiagnoses and improve the precision of the LLM diagnostic in neuroradiology. More research is needed to optimize these approaches for broader clinical implementation, balancing accuracy and utility.

Identifiants

pubmed: 39061677
pii: diagnostics14141541
doi: 10.3390/diagnostics14141541
pii:
doi:

Types de publication

Journal Article

Langues

eng

Subventions

Organisme : Japan Society for the Promotion of Science
ID : 22K07674

Auteurs

Akihiko Wada (A)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Toshiaki Akashi (T)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

George Shih (G)

Clinical Radiology, Weill Cornell Medical College, New York, NY 10065, USA.

Akifumi Hagiwara (A)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Mitsuo Nishizawa (M)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Yayoi Hayakawa (Y)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Junko Kikuta (J)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Keigo Shimoji (K)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Katsuhiro Sano (K)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Koji Kamagata (K)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Atsushi Nakanishi (A)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Shigeki Aoki (S)

Department of Radiology, Juntendo University Graduate School of Medicine, Tokyo 113-8421, Japan.

Classifications MeSH