Recent breakthroughs in artificial intelligence show that GPT-5 could revolutionize AI esophageal cancer staging by matching the diagnostic accuracy of human specialists. Accurate staging for esophageal squamous cell carcinoma traditionally relies on the complex interpretation of F-fluorodeoxyglucose positron emission tomography (F-FDG-PET) images. However, significant workforce shortages in radiology and surgery often delay these critical reports. Consequently, researchers have turned to large language models (LLMs) to automate and accelerate this high-stakes diagnostic process.

Improving Accuracy with AI Esophageal Cancer Staging

A retrospective study at Tohoku University Hospital evaluated the performance of six LLMs against four blinded human evaluators. The results demonstrated that GPT-5 achieved a patient-level diagnostic accuracy of 85.8%, which statistically matched the 85.0% accuracy of a nuclear medicine specialist. Furthermore, GPT-5 significantly outperformed radiology residents and previous model iterations like GPT-4 Turbo. This indicates that advanced models can handle complex visual-medical tasks with high precision. Specifically, the model assessed lymph node metastases and distant metastases using standardized maximum-intensity projection (MIP) images.

Moreover, the researchers utilized the Matthews Correlation Coefficient to ensure robust performance metrics despite potential class imbalances in the dataset. GPT-5 recorded an MCC of 0.704, suggesting a strong correlation between the model's predictions and the reference standard radiology reports. Similarly, the study highlighted the benefits of zero-shot model analysis, which requires no task-specific training. Therefore, these findings suggest that LLMs could soon serve as reliable automated support systems in oncology centers. Nevertheless, clinicians must continue to verify AI outputs against local guidelines to ensure patient safety.

Clinical Implications for Resource-Limited Settings

In countries like India, where esophageal cancer remains a major health challenge, such automated tools could alleviate the burden on tertiary care centers. For instance, many patients in rural areas present with advanced-stage disease, necessitating rapid and accurate staging for immediate intervention. Additionally, integrating AI into the workflow could help standardize reporting across different hospital levels. Although human oversight remains mandatory, the ability of GPT-5 to match specialist-level interpretation offers a promising solution to diagnostic delays.

FAQs

Can GPT-5 accurately stage esophageal cancer without human intervention?

While GPT-5 demonstrates accuracy comparable to nuclear medicine specialists, it is currently designed as a decision-support tool. Specialists should still review AI-generated staging to ensure clinical context and patient safety.

What type of imaging data does the model analyze?

The model analyzes frontal maximum-intensity projection (MIP) images from FDG-PET scans. It evaluates these along with tumor location data to determine clinical N and M stages.

How does GPT-5 compare to previous AI models in oncology?

GPT-5 significantly outperforms GPT-4 Turbo and earlier versions in diagnostic accuracy and Matthews Correlation Coefficient. This progress reflects improved medical reasoning and better processing of medical imaging data.

Disclaimer: This content is for informational and educational purposes only. It is not intended as a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or other qualified health provider with any questions you may have regarding a medical condition. Refer to the latest local and national guidelines for clinical practice.

References

Maruyama H et al. Evaluation of GPT-5 for Esophageal Cancer Staging Using Fluorodeoxyglucose Positron Emission Tomography Maximum-Intensity Projection Images: Comparative Pilot Study. JMIR Cancer. 2026 Feb 23. doi: 10.2196/86630. PMID: 41729569.

Talukdar B, Sharma B. Epidemiological Trends and Clinical Characteristics of Esophageal Cancer in North-East India: A Hospital-Based Descriptive Study from a Tertiary Cancer Center. Journal of Advances in Medicine and Medical Research. 2025 Nov 27.

Bhayana R et al. Large Language Models in Cancer Imaging: Applications and Future Perspectives. PMC. 2025 May 08.