Search Results
Found 2 results
510(k) Data Aggregation
(245 days)
AI-Rad Companion (Cardiovascular)
AI-Rad Companion (Cardiovascular) is image processing software that provides quantitative and qualitative analysis from previously acquired Computed Tomography DICOM images to support radiologists and physicians from emergency medicine, specialty care, urgent care, and general practice in the evaluation and assessment of cardiovascular diseases.
It provides the following functionality:
- Segmentation and volume measurement of the heart
- Quantification of the total calcium volume in the coronary arteries
- Segmentation of the aorta
- Measurement of maximum diameters of the aorta at typical landmarks
- Threshold-based highlighting of enlarged diameters
The software has been validated for non-cardiac chest CT data with filtered backprojection reconstruction from Siemens Healthineers, GE Healthcare, Philips, and Toshiba/Canon. Additionally, the calcium detection feature has been validated on non-cardiac chest CT data with iterative reconstruction from Siemens Healthineers.
Only DICOM images of adult patients are considered to be valid input.
AI-Rad Companion (Cardiovascular) SW version VA20 is an enhancement to the previously cleared device AI-Rad Companion (Cardiovascular) K183268 that utilizes machine and deep learning algorithms to provide quantitative and qualitative analysis to computed tomography DICOM images to support qualified clinicians in the evaluation and assessment of cardiovascular diseases.
As an update to the previously cleared device, the following modifications have been made:
Segmentation of Aorta – Performance Improvement
Although the structure of the underlying neural network has not changed in the subject device of this submission, the performance was enhanced over the previously cleared device by adding training data (re-use of existing annotations + 267 additional annotations).
Aorta diameter measurements - Maximum Diameter Ascending, Descending Aorta
In the previously cleared device diameter measurements of the aorta were performed at nine predefined locations according to the AHA guidelines.
As an enhancement to the previously cleared device and subject of this submission are aorta diameter measurements at the locations of the maximum diameter of the ascending and the descending aorta.
Visualization of aorta's VRT and as cross-sectional MPRs - Maximum Diameter Ascending, Descending Aorta
In the previously cleared device visualization VRT and cross-sectional MPRs were provided at nine predefined locations according to the AHA guidelines.
As an enhancement to the previously cleared device, such visualization of the maximum diameter of the ascending and descending aorta were added to the subject of this submission.
Categorization of diameter measurements - Maximum Diameter Ascending, Descending Aorta
In the previously cleared device categorization of diameter measurements was performed at locations according to the AHA guidelines.
With the subject of this submission, the categorization of diameter measurements was extended to locations of the maximum diameter of the ascending and descending aorta.
Individual Confirmation of Aorta Findings
For the measurements of the aorta, only all the measurements could be accepted or declined in the predicate device.
Within the scope of this submission the concept of individual accept, decline-possibility was introduced to all aorta measurements.
Structured DICOM Report (DICOM TID 1500)
In the predicate device, the system would produce results in form of quantitative, structured and textual reports and would generate DICOM Secondary Capture images which would be forwarded to PACS reading and reporting systems.
Within the scope of this submission, the system supports an alternative, digital output format for the same results. For this purpose, a DICOM Structured Report is generated which is both human and machine readable and, therefore, will support, e.g., a transfer of the results into the clinical report more efficiently. The DICOM Structured Report is compliant to the TID1500 format for applicable content.
Cloud and Edge Deployment
Another enhancement provided within this submission is the existing cloud deployment in an on-premise deployment known as an edge deployment. The system remains hosted in the teamplay digital health platform and remains driven by the AI-Rad Companion Engine; however, with the edge deployment the processing of clinical data and the generation of results is performed within the customer environment. This system remains fully connected to the cloud for monitoring and maintenance of the system from a remote setup. At the time of this submission this feature has been cleared in submission K213706 (AI-Rad Companion Brain MR VA40) and is unchanged within this subject device.
Here's a breakdown of the acceptance criteria and study details for the AI-Rad Companion (Cardiovascular) from the provided text:
1. Table of Acceptance Criteria and Reported Device Performance
| Feature/Metric | Acceptance Criteria (Predicate Device Performance) | Reported Device Performance (Subject Device) |
|---|---|---|
| Aorta Segmentation (DICE coefficient) | Mean DICE coefficient of 0.910 (± 0.066) | Mean DICE coefficient of 0.924 (± 0.046) |
| Aorta Diameter Measurements (9 predefined landmarks) - Bias | Bias within ±1.8 mm (95%-CI: [1.5 mm, 2.1 mm]) | Bias within ±1.5 mm (95%-CI: [0.9 mm, 2.0 mm]) |
| Aorta Diameter Measurements (9 predefined landmarks) - Mean Absolute Error (MAE) | MAE ≤2.4 mm (95%-CI: [2.1 mm, 2.6 mm]) | MAE ≤2.2 mm (95%-CI: [1.8 mm, 2.6 mm]) |
| Aorta Diameter Measurements (Max Ascending/Descending) - Percentage within Inter-Reader LoA | Inter-reader variability 95%-limits of agreement (LoAs) established at ±3.51 mm | 91.9% of measurements within LoA |
| Aorta Diameter Measurements (Max Ascending/Descending) - Bias | Not explicitly stated as acceptance criteria, but inter-reader variability was assessed. | Bias within ±1.5 mm (95%-CI: [1.2 mm, 1.8 mm]) |
| Aorta Diameter Measurements (Max Ascending/Descending) - MAE | Not explicitly stated as acceptance criteria, but inter-reader variability was assessed. | MAE ≤1.8 mm (95%-CI: [1.44 mm, 2.23 mm]) |
2. Sample Size and Data Provenance for Test Set
- Aorta Segmentation:
- Sample Size: N=315
- Data Provenance: Retrospective clinical cohort. Details regarding country of origin are not specified.
- Aorta Diameter Measurements:
- Sample Size: N=193
- Data Provenance: Representative retrospective clinical cohort. This included:
- Consecutive patients undergoing Chest CT exams for varying indications.
- A cohort at increased risk for incidental findings, particularly in the cardiovascular domain, due to the screening nature of the examination.
- Specific percentages: 50% of cases with dilated aorta, 9% of cases with aortic aneurysm.
- Details regarding country of origin are not specified.
3. Number and Qualifications of Experts for Test Set Ground Truth
The document does not explicitly state the number or qualifications of experts used to establish the ground truth for the test set. It mentions "inter-reader variability was assessed" for the diameter measurements, implying human expert involvement in establishing reference values, but provides no further details on their credentials or number.
4. Adjudication Method for the Test Set
The adjudication method is not explicitly stated in the provided text.
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
No Multi-Reader Multi-Case (MRMC) comparative effectiveness study (AI vs. human readers with AI assistance vs. without AI assistance) is mentioned. The study focuses on comparing the subject device's performance to the predicate device's performance, and algorithm accuracy against ground truth.
6. Standalone (Algorithm Only) Performance Study
Yes, standalone performance studies were done. The reported performance metrics (DICE coefficient, bias, MAE) are directly attributed to the device's algorithms in comparison to established (likely expert-defined) ground truth, without human interaction during the measurement process. The text explicitly states, "the performance of the aorta segmentation module has been validated... For the subject device, average DICE (± std. dev) coefficient was 0.924 (± 0.046)." Similarly for the diameter measurements, "91.9% of the measurements provided by the subject device were found to lie within the LoA," indicating standalone algorithmic measurement.
7. Type of Ground Truth Used for Test Set
The type of ground truth used is implied to be expert consensus or expert measurements, especially for the aorta diameter measurements where "inter-reader variability was assessed" and the device's measurements were compared against these established ranges. For segmentation, the DICE coefficient comparison against the predicate suggests a reference standard, likely also based on expert annotations or consensus.
8. Sample Size for the Training Set
The document notes that the performance of the aorta segmentation was "enhanced over the previously cleared device by adding training data (re-use of existing annotations + 267 additional annotations)." This indicates that the training set for the aorta segmentation was augmented with at least 267 new cases, in addition to previously used annotations from the predicate device's training. The total size of the training set is not explicitly given, only the additional annotations for the updated model.
9. How the Ground Truth for the Training Set was Established
The text states "re-use of existing annotations + 267 additional annotations" for the training data. While it doesn't explicitly detail how these annotations were created, "annotations" typically refer to expert-labeled segmentation masks or measurements. This implies that medical professionals (e.g., radiologists) meticulously outlined structures or took measurements on the training images to serve as the ground truth for training the deep learning model.
Ask a specific question about this device
(291 days)
AI-Rad Companion (Cardiovascular)
AI-Rad Companion (Cardiovascular) is image processing software that provides quantitative and qualitative analysis from previously acquired Computed Tomography DICOM images to support radiologists and physicians from emergency medicine, specialty care, urgent care, and general practice in the evaluation and assessment of cardiovascular diseases.
It provides the following functionality:
- · Segmentation and volume measurement of the heart
- · Quantification of the total calcium volume in the coronary arteries
- Segmentation of the aorta
- · Measurement of maximum diameters of the aorta at typical landmarks
- · Threshold-based highlighting of enlarged diameters
The software has been validated for non-cardiac chest CT data with filtered backprojection reconstruction from Siemens Healthineers, GE Healthcare, Philips, and Toshiba/Canon. Additionally, the calcium detection feature has been validated on non-cardiac chest CT data with iterative reconstruction from Siemens Healthineers.
Only DICOM images of adult patients are considered to be valid input.
In general, AI-Rad Companion (Cardiovascular) is a software only image post-processing application that uses deep learning algorithms to post-process CT data of the thorax.
The subject device AI-Rad Companion (Cardiovascular) is an image processing software that utilizes deep learning algorithms to provide quantitative and qualitative analysis from previously acquired Computed Tomography DICOM images to support radiologists and physicians from emergency medicine, specialty care, urgent care, and general practice in the evaluation and assessment of disease of the thorax. The subject device supports the following device specific functionality:
- Segmentation and volume measurement of heart
- . Identification and measurement of volume with high Hounsfield values -- related to coronary calcification
- . Segmentation of the aorta and determination of 9 Landmarks
- . Computation of cross-sectional MPRs at the 9 landmarks and their maximum diameter
- Measurement of maximum diameters of the aorta at typical landmarks ●
- Threshold-based classification of diameters into different categories ●
Here's an analysis of the acceptance criteria and study details for the AI-Rad Companion (Cardiovascular) device, based on the provided text:
1. Table of Acceptance Criteria and Reported Device Performance
The document doesn't explicitly present a formal "acceptance criteria" table with specific pass/fail thresholds for each metric. Instead, it reports performance results (sometimes implicitly as "equivalent" or "consistent") that are presumably within acceptable limits for FDA clearance given the substantial equivalence claim.
| Feature / Metric | Acceptance Criteria (Implicitly Met) | Reported Device Performance and Confidence Intervals |
|---|---|---|
| Coronary Calcium Volume Quantification | Performance equivalent to predicate device | Logarithmic correlation coefficient of total coronary calcium volume between subject and predicate device was 0.96 (N=381). |
| Aorta Diameter Measurements (Average Absolute Error) | Performance consistent across critical subgroups and within acceptable limits | Average absolute error in aorta diameters was 1.6 mm (95% confidence interval: [1.5 mm, 1.7 mm]) across all nine measurement locations. |
| Aorta Diameter Measurements (Per Location) | Performance consistent across critical subgroups and within acceptable limits | Varied between 0.9 mm and 2.4 mm per location (N=193). |
| Consistency across Subgroups | Performance consistent for critical subgroups (vendors, slice thickness) | Performance was consistent for all critical subgroups, such as vendors or slice thickness. |
| Software Functionality | All software specifications met acceptance criteria | All testable requirements in the Engineering Requirements Specifications keys, Subsystem Requirements Specifications keys, and the Risk Management Hazard keys have been successfully verified and traced. Testing results support that all software specifications have met the acceptance criteria. |
| Risk Management | Identified hazards are mitigated | Risk analysis completed and risk control implemented to mitigate identified hazards. |
| Human Factors Usability | Human factors addressed and acceptable for safe and effective use | Human Factor Usability Validation showed that Human factors are addressed in the system test and in clinical use tests with customer reports and feedback. |
2. Sample Sizes and Data Provenance
- Test Set Sample Sizes:
- Coronary Calcium Volume: N = 381 data sets.
- Aorta Diameter Measurements: N = 193 data sets.
- Data Provenance: Retrospective performance studies on non-cardiac chest CT data from multiple clinical sites across the United States.
3. Number of Experts and Qualifications
The document does not explicitly state the "number of experts" or their specific "qualifications" used to establish ground truth. However, for the training set, it mentions "Description of ground truth / annotations generation," implying expert involvement. For the validation set, the comparison is primarily against a "predicate device," suggesting that the ground truth for that comparison would have been established previously for the predicate, not necessarily by new experts for this study.
4. Adjudication Method for the Test Set
The document does not specify any adjudication method (e.g., 2+1, 3+1) for the test set's ground truth. The comparison seems to be against the predicate device's output, which would have its own established ground truth based on its clearance.
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
- No, an MRMC comparative effectiveness study was not explicitly described in the provided text. The study focuses on the standalone performance of the AI device in comparison to a predicate device, not on how human readers improve with or without AI assistance.
6. Standalone Performance Study (Algorithm Only)
- Yes, a standalone performance study was done. The document states: "The performance of the AI-Rad Companion (Cardiovascular) device has been validated in retrospective performance studies..." and details the algorithm's performance metrics (correlation coefficient for calcium, absolute error for aorta diameters) against "predicate device" or implied ground truth, indicating algorithm-only performance.
7. Type of Ground Truth Used
The ground truth for the validation studies appears to be based on:
- Comparison to a predicate device's output: For coronary calcium volume quantification, the performance is reported as a correlation between the subject device and the predicate device.
- Likely expert-derived measurements previously established or derived from the predicate device's method: For aorta diameter measurements, the "average absolute error" suggests comparison against a reference "true" measurement, which would typically be derived by experts using the predicate's methodology, or a gold standard measurement. The mention of "AHA standard" for diameter categorization also points to established clinical guidelines as a reference.
8. Sample Size for the Training Set
The document states: "Training cohort: size and properties of data used for training O". However, it does not explicitly provide the numerical sample size for the training set.
9. How the Ground Truth for the Training Set Was Established
The document briefly mentions under "Data" for each algorithm analysis: "Description of ground truth / annotations generation O". This implies that ground truth was established, likely through expert annotation or a similar process, but does not provide specific details on the methodology (e.g., number of annotators, their qualifications, consensus process) for the training set's ground truth.
Ask a specific question about this device
Page 1 of 1