INTRODUCTION: Gynecologic patients with acute abdominal pain frequently present with cul-de-sac fluid. Differentiating hemorrhagic from non-hemorrhagic fluid is critical, as hemoperitoneum requires urgent surgical intervention. This distinction is particularly challenging when the imaging characteristics of the fluid are inconclusive in distinguishing between hemorrhagic and non-hemorrhagic types, especially in time-sensitive settings. METHODS: We developed the CLEAR (Cul-de-sac Liquid Evaluation And Recognition) system, a deep learning model based on convolutional neural networks. It was trained on 2,200 transvaginal ultrasound and CT images from 100 surgical cases at Ilsim Medical Foundation Woori Hospital, South Korea. Ground truth labels were derived from laparoscopic findings. CLEAR was trained over 100,000 iterations to analyze echogenicity and attenuation, enabling classification of fluid as hemorrhagic, infectious, or other types. RESULTS: CLEAR demonstrated high diagnostic accuracy, particularly in identifying hemorrhagic fluid when conventional imaging was inconclusive. Multimodal integration of ultrasound and CT significantly enhanced diagnostic performance. The model provided real-time classification and maintained consistent accuracy across diverse patient demographics. CONCLUSIONS/IMPLICATIONS: CLEAR enables rapid and accurate interpretation of cul-de-sac fluid using standard imaging modalities. It facilitates timely clinical decision-making in ambiguous cases, such as when distinguishing between hemorrhagic and non-hemorrhagic fluid is challenging. By enhancing diagnostic confidence and consistency, CLEAR may contribute to improved patient outcomes. Further multicenter validation is warranted to confirm its broader clinical applicability.
Cheon et al. (Thu,) studied this question.