FDA 510(k) Search - By Innolitics (SaMD and AI Experts)

The Pixel 3 System is intended to be used with surgical sponges, software, hardware and accessory devices which have been validated for use with the Pixel 3 System to estimate the hemoglobin (Hb) mass contained on used surgical sponges. The Pixel 3 System is also intended to calculate an estimate of blood volume on used surgical sponges from the estimated Hb mass and a user-entered patient serum Hb value. The validated surgical sponges, hardware, software, accessory devices and Hb mass ranges are listed in the Instructions for Use.

The Pixel 3 System is also indicated for use to aid in counting surgical sponges and may be used to record and display case-specific blood components infused over time. The Pixel 3 System is additionally indicated for use to aid in managing surgical sponges, including providing a visual record of sponge images, and to record the user-entered weight of used surgical sponges in order to calculate an estimate of fluid volume on the sponges.

Device Description

The Pixel 3 System is a software program (mobile medical application) used on an Apple iPad® tablet to capture images of used surgical sponges to assist surgical personnel in the management of surgical sponges after surgical use and to aid in the estimation of blood loss. The main functions of the device are summarized below.

The Pixel 3 System provides an estimate of the Hb mass lost onto the sponge (sHbL), which is derived from a software algorithm that analyzes images of sponges sent to the Gauss off-site server along with user-entered information about the type of sponge. An estimate of the cumulative blood volume lost onto the sponges (sEBL) is subsequently calculated by dividing the sHbL for each sponge by a user-entered value for the patient's laboratory-derived serum Hb level at the time of image capture. Whereas sHbL is estimated independently from the laboratory-derived serum Hb (i.e., directly from each image), sEBL is derived from a calculator whose inputs are adjustable by the user. The Pixel 3 System provides this estimate of blood content on sponges (i.e., sEBL by sHbL method) and estimate of sHbL only for the validated laparotomy sponge types listed in Table 1.

The Pixel 3 System may also be used to track the weight of soaked surgical sponges recorded by the user. The device may aid in the estimation of blood loss by calculating an estimate of the cumulative sEBL by weight, provided that a dry and wet weight has been entered for each sponge. This estimate (i.e., sEBL by weight method) is based on the total weight of the soaked sponges less their dry weights normalized by the density of whole blood (1.060 g/mL). For the sEBL by weight method, a user may manually enter sponge types other than those validated for the sEBL by sHbL method (see Table 1 above); however, those sponge types can only be used to calculate sEBL by weight.

The Pixel 3 System also allows surgical personnel to categorize sponges by sponge type and provides an automated ongoing count of the total number of sponge images and sponge images by tag. The device allows for the input and display of case-specific values pertaining to fluid management during surgical procedures (e.g., packed red blood cell volume administered over time, fresh frozen plasma volume administered over time, platelet volume administered over time), as detailed in the Instructions for Use. The Pixel 3 System also provides a visual record of images for further evaluation during the surgical case.

To use the device, the user mounts the iPad tablet onto an IV pole; the device contains alignment indicators to help the user align the IV pole. The user then places a sponge in view of the iPad camera (the device contains a camera-bounding box to help the user with sponge placement), and scans an image of the sponge by touching the iPad screen or using an optional wireless foot pedal. The device also contains an ambient light indicator, which helps the user determine when a poor (indicator is yellow) or appropriate (indicator is white) level of ambient light is present for image capture.

The full-screen display of sHbL and sEBL by sHbL outputs includes an estimate of the cumulative error of the Pixel 3 System, computed as the 95% Bland-Altman Limits of Agreement (and denoted as "95% limits Bland Altman" on the display). The display of estimated error is updated on a real-time basis, as successive sponges are accumulated and scanned. A Bland-Altman plot in biostatistics is a method of data plotting used to analyze the agreement between two different assays. Bias is defined as the arithmetic mean of the differences between the device's output value and measurements obtained using a reference standard. The Bland- Altman Limits of Agreement represent two standard deviations (1.96 x SD) of the differences around the bias, and represent the error range within which 95% of all differences between the device's output and the reference standard's measures are expected to lie.

AI/ML Overview

Here's a summary of the acceptance criteria and the study that proves the Pixel 3 System meets them, based on the provided document:

Acceptance Criteria and Device Performance

Acceptance Criteria (Special Controls)	Reported Device Performance and Compliance
1. Non-clinical performance data must demonstrate that the device performs as intended under anticipated conditions of use, with comparison to a scientifically valid alternative method for measuring deposited hemoglobin mass. These include lighting, Hb concentrations, blood volume absorption, and presence of non-sanguineous fluids.	Bench Testing: The Pixel 3 System demonstrated a strong positive linear correlation (r = 0.92 [95% CI 0.91 to 0.93]) with the reference method (pre-measured Hb mass). The Hb mass bias was 0.01 g Hb with limits of agreement of approximately ± 1.2 g per sponge. It was tested across different sponge types, Hb concentrations (5-17 g/dl), fluid and blood volumes, and various ambient light conditions. The pre-determined acceptance criterion was (b)(4) Trade Secrete/CCI per sponge.
2. Human factors testing and analysis must validate that the device design and labeling are sufficient for appropriate use by intended users.	Human Factors Testing (SW 02089): Evaluated usability with registered nurses (circulating nurses). All 8 users in simulated cases successfully completed tasks. 94% of tasks completed on the first pass, 100% on the second. Likert scores for usability were high (average 4.57). No users identified safety issues.
3. Appropriate analysis and non-clinical testing must validate the electromagnetic compatibility (EMC) and wireless performance of the device.	EMC and Wireless Technology Testing: Compliant with IEC 60601-1-2:2007 (Class B) for EMC. Maintained essential wireless functionality (capturing sponge image and receiving sHbL within one minute) under various noisy wireless environments (single/multiple WiFi, Bluetooth, RFID interferers, maximum power).
4. Appropriate software verification, validation and hazard analysis must be performed.	Software Documentation: Adequately described software development, hazard analysis, and validation/verification activities. System-level test protocols, pass/fail criteria, and results were provided. Cybersecurity issues and mitigation were addressed.
5. Software display must include an estimate of the cumulative error associated with estimated blood loss values.	Device Description: The full-screen display of sHbL and sEBL outputs includes an estimate of the cumulative error, computed as the 95% Bland-Altman Limits of Agreement. This display is updated in real-time.
6. Labeling must include warnings, cautions, limitations, performance testing summary, validated materials, and EMC/wireless instructions.	Labeling: Provided with required warnings, cautions, and limitations (e.g., MR Unsafe, not a sterile device, not to be used as a trigger for clinical action). Includes a detailed summary of performance testing (bias, variance), validated surgical materials (sponge types), and EMC/wireless instructions.

Study Details

1. Sample sizes used for the test set and data provenance:

Bench Testing:
- Test Set (sponges): 50 sponges of each of the four validated types (NovaPlus, AMD Ritmed, RFDetect Premium, Allegiance Disposable).
- Total Scans: (b)(4) Trade Secret/CCI scans per sponge type. The exact total number of sponges in the test set isn't explicitly stated as a single number but would be 50 sponges/type * 4 types = 200 sponges. Each sponge was scanned multiple times.
- Data Provenance: Not explicitly stated, but likely laboratory-controlled settings (simulated blood loss).
Human Factors Testing:
- Subjects: Registered nurses (circulating nurses).
- Test Set: (b)(4) Trade Secret/CCI live surgical case evaluations and 8 simulated task-based evaluations.
- Data Provenance: Single-center study (location not specified), prospective.
Preliminary Clinical Testing (Study 1):
- Patients: 46 patients
- Sponges: 758 laparotomy sponges ([b]4] Trade Secret/CCI, RFDetect, and AMD Ritmed).
  - 167 sponges analyzed on a per-sponge basis.
  - 591 sponges analyzed in batches.
- Data Provenance: Prospective, multi-center (3 clinical sites, location not specified). Conducted between July and November 2012.
Confirmatory Clinical Testing (Study 2):
- Patients: 50 patients
- Sponges: 791 laparotomy sponges (RFDetect laparotomy sponges). All individually assayed.
- Data Provenance: Prospective, single-center (1 clinical site, location not specified). Enrollment from July 2013 to October 2013.

2. Number of experts used to establish the ground truth for the test set and their qualifications:

Bench Testing: The ground truth was established by pre-measured Hb mass deposited in controlled amounts. No human experts were used for this ground truth as it was a direct measurement.
Clinical Studies (Study 1 & 2): The ground truth for Hb mass was established using a "mechanical extraction method (assay sHbL)" or "rinsing and photometric assay of the effluent (reference method)." This indicates a laboratory method rather than expert human interpretation.

3. Adjudication method for the test set:

Bench Testing: Not applicable, as ground truth was direct measurement.
Clinical Studies (Study 1 & 2): Not applicable, as ground truth was established by a laboratory reference method.

4. If a multi-reader multi-case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance:

A formal MRMC study comparing AI-assisted human readers to unassisted human readers was not explicitly described in the provided text.
However, the Confirmatory Clinical Testing (Study 2) did compare the Pixel 3 System's performance (sEBL) to traditional human methods:
- Visual (single-rater) estimates by the anesthesiologist.
- Gravimetric (weighing sponges) methods.
Effect Size (Comparison to traditional methods): In Study 2, the Pixel 3 System's sEBL (end of case) displayed a consistently lower variance and higher precision than both visual estimates and gravimetric methods. The gravimetric method overestimated blood loss by 359 ml per patient compared to the Pixel 3 System in Study 1. This suggests significant improvement in accuracy and precision over traditional human-driven methods.

5. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done:

Yes, the Bench Testing was a standalone evaluation of the algorithm's accuracy in estimating Hb mass compared to a pre-measured reference.
The Clinical Studies (Study 1 and 2) also evaluated the Pixel 3 System's output (sHbL and sEBL) as a standalone estimate compared to the laboratory reference method, noting its bias and correlation. While users were present to scan, the performance metrics (bias, correlation, limits of agreement) directly reflect the algorithm's output against the ground truth.

6. The type of ground truth used:

Bench Testing: Direct measurement of pre-deposited Hb mass on sponges.
Clinical Studies (Study 1 & 2): Laboratory-based reference method involving "mechanical extraction" or "rinsing and photometric assay of the effluent" to measure Hb mass. This is a highly objective, quantifiable ground truth.

7. The sample size for the training set:

The document does not explicitly state the sample size for the training set. It focuses on the validation studies.

8. How the ground truth for the training set was established:

Since the training set size and details are not provided, information on how its ground truth was established is also not available in the given text. However, given the methodology of the test sets, it is highly probable that similar laboratory-based reference methods would have been used for training data ground truth.

Summary

Regulation Number and Section

§ 880.2750 Image processing device for estimation of external blood loss.

(a)
Identification. An image processing device for estimation of external blood loss is a device to be used as an aid in estimation of patient external blood loss. The device may include software and/or hardware that is used to process images capturing externally lost blood to estimate the hemoglobin mass and/or the blood volume present in the images.(b)
Classification. Class II (special controls). The special controls for this device are:(1) Non-clinical performance data must demonstrate that the device performs as intended under anticipated conditions of use. Demonstration of the performance characteristics must include a comparison to a scientifically valid alternative method for measuring deposited hemoglobin mass. The following use conditions must be tested:
(i) Lighting conditions;
(ii) Range of expected hemoglobin concentrations;
(iii) Range of expected blood volume absorption; and
(iv) Presence of other non-sanguineous fluids (
e.g., saline irrigation fluid).(2) Human factors testing and analysis must validate that the device design and labeling are sufficient for appropriate use by intended users of the device.
(3) Appropriate analysis and non-clinical testing must validate the electromagnetic compatibility (EMC) and wireless performance of the device.
(4) Appropriate software verification, validation, and hazard analysis must be performed.
(5) Software display must include an estimate of the cumulative error associated with estimated blood loss values.
(6) Labeling must include:
(i) Warnings, cautions, and limitations needed for safe use of the device;
(ii) A detailed summary of the performance testing pertinent to use of the device, including a description of the bias and variance the device exhibited during testing;
(iii) The validated surgical materials, range of hemoglobin mass, software, hardware, and accessories that the device is intended to be used with; and
(iv) EMC and wireless technology instructions and information.