From Bedside to Bot-Side: Artificial Intelligence in Emergency Appendicitis Management.

Area of Science:

• Medical Artificial Intelligence
• Surgical Decision Support Systems
• Gastrointestinal Surgery

Background:

• Acute appendicitis (AA) is a common surgical emergency with potential for severe complications.
• Accurate diagnosis and timely treatment are crucial for managing AA and preventing morbidity/mortality.
• Current decision-making relies on clinical, laboratory, and radiological data, with surgical intervention often being the primary approach.

Purpose of the Study:

• To compare the diagnostic and treatment recommendation accuracy of GPT-4.5, DeepSeek R1, and a machine learning model against board-certified surgeons for acute appendicitis.
• To evaluate the impact of incorporating clinical guidelines on AI model performance.
• To assess the potential of AI in assisting surgical decision-making for patients with lower abdominal pain.

Main Methods:

• A multicenter retrospective study included 63 confirmed appendicitis cases and 50 controls.
• Clinical, laboratory, and radiological data were used to train and test AI models (GPT-4.5, DeepSeek R1, Random Forest ML).
• AI recommendations for appendectomy versus conservative antibiotic therapy were compared against a reference standard of board-certified surgeons.

Main Results:

• DeepSeek R1 showed non-significant accuracy improvement from 80.5% to 83.2% with guideline integration.
• GPT-4.5 accuracy increased from 70.8% to 76.1% with guideline integration.
• The machine learning model achieved 84.3% training accuracy and 85.0% validation accuracy.

Conclusions:

• GPT-4.5, DeepSeek R1, and machine learning models show potential as adjuncts for surgical decision-making in appendicitis.
• These AI tools may be particularly beneficial in resource-limited environments.
• Continued research and validation are necessary to refine AI model performance for clinical application.