EdgeFlow UH10

The EdgeFlow UH10 is a portable ultrasound device that employs a deep learning model for the noninvasive measurement of bladder urine volume. Its methodology relies on AI-driven B-mode ultrasound image classification for bladder detection and semantic segmentation for boundary delineation, with clinical validation demonstrating high accuracy through F1 and Dice scores.

The EdgeFlow UH10 is a portable ultrasound system designed for noninvasive measurement of urine volume in the bladder. It comprises a handheld console with a 3.2-inch touchscreen display and an integrated probe, utilizing B-mode and harmonic imaging modes. The core of its operation is a deep learning model which includes feature extraction, a binary classification network to determine bladder presence, and a semantic segmentation network to delineate the bladder area from ultrasound images. The system calculates bladder volume based on these AI-generated contours, displays the images and volume, and allows for data export via USB or Wi-Fi. It is powered by a rechargeable lithium-ion battery. The performance of the EdgeFlow UH10's deep learning model was validated using clinical data collected under an Institutional Review Board-approved trial, with distinct datasets for training and testing. For the classification network, tested on 3,711 images from 90 subjects, the device achieved an F1 score of 0.979 (95% CI 0.974–0.984), exceeding the predefined success criterion of ≥0.90, and also met its secondary endpoint criterion for Precision-Recall Area Under Curve (PR AUC) of ≥0.95. The segmentation network, evaluated using 1,528 images, achieved a Dice Score of 0.896 (95% CI 0.890–0.901), meeting its performance criterion of ≥0.89. These results demonstrate the system's accuracy for its intended use by qualified healthcare professionals in professional healthcare facilities.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the EdgeFlow UH10's deep learning model account for anatomical variability in bladder shape across different patient demographics (e.g., age, sex, BMI) beyond the collected auxiliary demographic references?; What specific features are extracted by the 'feature extraction' component of the deep learning model, and how critical are they for the subsequent classification and segmentation tasks?; The document states the Precision-Recall Area Under Curve (PR AUC) criterion for the classification network was ≥0.95 and was met; what was the exact achieved PR AUC value, and how does it compare to the F1 score in reflecting performance on imbalanced datasets?; What were the inter-rater reliability metrics for the two evaluators who established the ground truths for the test dataset, and how were discrepancies in their reviews resolved?; How does the device's measurement accuracy of ±7.5mL for 0-100mL and ±7.5% for 100-999mL specifically perform at the extreme ends of these ranges, particularly for very low or very high bladder volumes?