Search Results
Found 1 results
510(k) Data Aggregation
(271 days)
The EdgeFlow UH10 is an ultrasound device intended to be used for measuring the urine volume in the bladder noninvasively. It is intended for use in professional healthcare facilities, such as hospitals, clinics, by qualified and trained healthcare professionals. The EdgeFlow UH10 supports B-mode and harmonic imaging modes.
This device is an ultrasound system that measures urinary bladder volume noninvasively using ultrasound. The system consists of a handheld unit (console) with a 3.2inch touchscreen display and a permanently attached probe. The system calculates the bladder volume based on ultrasound images using the probe. The images and the bladder volume are displayed on the screen. The results of the exam are automatically saved and saved exams are managed on the screen. Data can be exported to a standard PC using a wired USB export or Wi-Fi communications.
The system includes a rechargeable lithium-ion battery can be charged in the system directly connecting a USB cable with a power adapter. Also, a cradle is available for charging the battery or printing the results.
The deep learning model employed in EdgeFlow UH10 comprises three components: feature extraction, binary classification, and semantic segmentation networks. B-mode ultrasound images undergo classification in the network to determine the presence of the bladder, while the segmentation network is responsible for delineating the bladder area. Live B-mode ultrasound images are acquired once the scan button is pressed to start scanning. The output of the deep learning model manifests as the bladder contours displayed as green lines in the ultrasound images, and the bladder volume is subsequently calculated based on these lines when the scan button is pressed.
The EdgeFlow UH10 is an ultrasound device intended for non-invasive bladder urine volume measurement. Its performance was evaluated against specific acceptance criteria for its deep learning model, which comprises a classification network for bladder presence detection and a segmentation network for bladder delineation.
Here's a breakdown of the acceptance criteria and study details:
1. Acceptance Criteria and Reported Device Performance
| Component/Metric | Acceptance Criteria | Reported Device Performance (95% CI) |
|---|---|---|
| Classification Network | F1 score: ≥ 0.90 | 0.979 (0.974-0.984) |
| PR AUC: ≥ 0.95 (Secondary Endpoint) | Not explicitly reported in the text, but stated as satisfied | |
| Segmentation Network | Dice Score: ≥ 0.89 | 0.896 (0.890-0.901) |
2. Sample Size and Data Provenance
- Test Set Sample Sizes:
- Classification Network: 3,711 images
- Segmentation Network: 1,528 images
- Data Provenance: Data was collected under a clinical trial approved by Yonsei University Institutional Review Board in Severance Hospital (IRB No: 1-2022-0076). This indicates the data is prospective and originates from South Korea.
3. Number of Experts and Qualifications for Ground Truth Establishment (Test Set)
- Number of Experts: Two evaluators.
- Qualifications of Experts: The evaluators have "clinical experiences." Specific details on years of experience or precise professional titles (e.g., radiologist) are not provided in the document.
4. Adjudication Method for the Test Set
The document states that the "test dataset is independently reviewed by two evaluators... The review results are transformed into ground truths." This suggests a process where the two independent reviews were combined or reconciled to establish the final ground truth. It does not explicitly mention a specific adjudication method like 2+1 or 3+1, but the "independently reviewed" followed by "transformed into ground truths" implies some form of consensus or reconciliation.
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
A multi-reader multi-case (MRMC) comparative effectiveness study was not conducted. The reported studies focus on the standalone performance of the AI algorithm. There is no mention of human readers improving with or without AI assistance.
6. Standalone (Algorithm Only) Performance
Yes, standalone (algorithm only without human-in-the-loop performance) evaluations were conducted for both the classification and segmentation networks against established ground truths. The reported F1 score and Dice score are measures of the algorithm's direct performance.
7. Type of Ground Truth Used
The ground truth used for the test set was established through independent review by two evaluators with clinical experience. This indicates an expert consensus or expert-derived ground truth.
8. Sample Size for the Training Set
- Classification Network: 9,422
- Segmentation Network: 7,115
9. How the Ground Truth for the Training Set Was Established
The document states that "All data used for training and test dataset were collected under a clinical trial approved by Institutional Review Board with training data being independent from the test data." While it details how the test set ground truth was established by two independent evaluators, it does not explicitly describe the method for establishing the ground truth for the training set. It can be inferred that a similar process of expert review under IRB approval was likely employed, given the overall rigor of the study design.
Ask a specific question about this device
Page 1 of 1