Use of Automated Machine Learning for Classifying Hemoperitoneum on Ultrasonographic Images of Morrison's Pouch: A...

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Emergency medicine diagnostics

Background:

• Hemoperitoneum is a critical finding in trauma patients.
• Ultrasonography (USG) is a key diagnostic tool for detecting hemoperitoneum.
• Accurate and rapid classification of hemoperitoneum is vital for timely treatment.

Purpose of the Study:

• To evaluate the efficacy of automated machine learning (AutoML) in classifying hemoperitoneum.
• To assess the performance of AutoML using ultrasonography (USG) images of Morrison's pouch.
• To validate the diagnostic accuracy of AutoML in a real-world trauma setting.

Main Methods:

• A multicenter retrospective study involving 864 trauma patients.
• Training and internal validation of Google's open-source AutoML on 2000 USG images.
• External validation using a separate set of 200 USG images from a different trauma center.

Main Results:

• Internal validation demonstrated 95% sensitivity, 99% specificity, and an AUROC of 0.97.
• External validation achieved 94% sensitivity, 99% specificity, and an AUROC of 0.97.
• No statistically significant difference in performance between internal and external validation (p=0.78).

Conclusions:

• Publicly available AutoML accurately classifies hemoperitoneum in USG images of Morrison's pouch.
• AutoML offers a reliable tool for diagnosing hemoperitoneum in emergency trauma care.
• The findings support the integration of AutoML into clinical diagnostic workflows for trauma patients.