Critical Care Suite with Pneumothorax Detection AI Algorithm is a computer-aided triage, notification, and diagnostic device that analyzes frontal chest X-ray images for the presence of a pneumothorax. Critical Care Suite identifies and highlights images with a pneumothorax to enable case prioritization or triage and assist as a concurrent reading aide during interpretation of radiographs.

Intended users include qualified independently licensed healthcare professionals (HCPs) trained to independently assess the presence of pneumothoraxes in radiographic images and radiologists.

Critical Care Suite should not be used in-lieu of full patient evaluation or solely relied upon to make or confirm a diagnosis. It is not intended to replace the review of the X-ray image by a qualified physician. Critical Care Suite is indicated for adults and Transitional Adolescents (18 to < 22 years old but treated like adults).

Device Description

Critical Care Suite is a suite of Al algorithms for the automated image analysis of frontal chest X-rays acquired on a digital x-ray system for the presence of critical findings. Critical Care Suite with Pneumothorax Detection Al Algorithm is indicated for adults and transitional adolescents (18 to <22 years old but treated like adults) and is intended to be used by licensed qualified healthcare professionals (HCPs) trained to independently assess the presence of pneumothoraxes in radiographic images and radiologists. Critical Care Suite is a software module that can be deployed on several computing platforms such as PACS, On Premise, On Cloud or X-ray Imaging Systems.

Today's clinical workflow, hospitals are overburdened by large volume of orders and long turnaround times for radiologist reports. Critical Care Suite with the Pneumothorax Detection Al Algorithm enables effective prioritization and assists in the detection / diagnosis of pneumothoraxes for radiologists and HCPs that have been trained to independently assess the presence of pneumothoraxes in radiographic images. It performs this task by flagging images with a suspicious finding and providing a localization overlay of the suspected pneumothorax as well as a graphical representation of the algorithm's confidence in the resultant finding. These outputs can be displayed wherever the reviewing physician normally conducts their reads per their standard of care, including PACS, On Premise, On Cloud and Digital Projection Radiographic Systems.

AI/ML Overview

Here's a summary of the acceptance criteria and study details for the GE Medical Systems, LLC Critical Care Suite with Pneumothorax Detection AI Algorithm, based on the provided document:

1. Table of Acceptance Criteria and Reported Device Performance

The document primarily focuses on reporting the device's performance against its own established criteria rather than explicitly listing pre-defined "acceptance criteria" tables. However, we can infer the acceptance criteria from the reported performance goals.

Metric	Acceptance Criteria (Implied from Performance)	Reported Device Performance (Standalone)	Reported Device Performance (MRMC with AI Assistance vs. Non-Aided)
Pneumothorax Detection (Standalone Algorithm)	Detect pneumothorax in frontal chest X-ray images, with high diagnostic accuracy.	AUC of 96.1% (94.9%, 97.2%)	Not Applicable
Sensitivity (Overall)	High sensitivity for overall pneumothorax detection.	84.3% (80.6%, 88.0%)	Not Applicable
Specificity (Overall)	High specificity for overall pneumothorax detection.	93.2% (90.8%, 95.6%)	Not Applicable
Sensitivity (Large Pneumothorax)	High sensitivity for large pneumothoraxes.	96.3% (93.1%, 99.2%)	Not Applicable
Sensitivity (Small Pneumothorax)	High sensitivity for small pneumothoraxes.	75.0% (69.2%, 80.8%)	Not Applicable
Pneumothorax Localization (Standalone Algorithm)	Localize suspected pneumothoraxes effectively.	Partially localized 98.1% (96.6%, 99.6%) of actual pneumothorax within an image (apical, lateral, inferior regions).	Not Applicable
	Full agreement between regions.	67.8% (62.7%, 73.0%)	Not Applicable
	Overlap with true pneumothorax area.	DICE Similarity Coefficient of 0.705 (0.683, 0.724)	Not Applicable
Reader Performance Improvement (MRMC Study)	Improve reader performance for pneumothorax detection.	Mean AUC improved by 14.5% (7.0%, 22.0%; p=.002) from 76.8% (non-aided) to 91.3% (aided).	14.5% improvement in mean AUC
Reader Sensitivity Improvement	Increase reader sensitivity.	Reader sensitivity increased by 16.3% (13.1%, 19.5%; p<.001) from 67.4% (non-aided) to 83.7% (aided).	16.3% improvement in sensitivity
Reader Specificity Improvement	Increase reader specificity.	Reader specificity increased by 12.4% (9.6%, 15.1%; p<.001) from 76.6% (non-aided) to 89.0% (aided).	12.4% improvement in specificity
Reader Performance Improvement (Large Pneumothorax)	Improve reader performance for large pneumothoraxes.	Mean AUC improved by 10.5% (3.2%, 17.8%, p=0.009). Sensitivity improved by 13.4% (10.0%, 16.9%, p<.001).	10.5% improvement in mean AUC (large); 13.4% improvement in sensitivity (large)
Reader Performance Improvement (Small Pneumothorax)	Improve reader performance for small pneumothoraxes.	Mean AUC improved by 17.6% (9.3%, 25.9%, p<0.001). Sensitivity improved by 18.7% (13.8%, 23.6%, p<.001).	17.6% improvement in mean AUC (small); 18.7% improvement in sensitivity (small)
Improvement Across User Groups	Demonstrate improvement across different clinical user types.	All physicians (Rad, IM, ER) improved 10.4% (2.8%, 17.9%, p=0.015). Nurse practitioners improved 24.1% (1.2%, 47.0%, p=0.045). Non-radiologists (ER, IM, NP) improved 17.5% (9.6%, 25.4%, p<0.001).	Varied improvements across user groups as noted.

2. Sample Size Used for the Test Set and Data Provenance

Sample Size for Test Set: 804 images
Data Provenance: The test set included images from two North American sites.
Retrospective/Prospective: The document does not explicitly state if the test set was retrospective or prospective. However, given it's a "final validation ground truth dataset" that was not used in training, it's highly likely to be retrospective.

3. Number of Experts Used to Establish the Ground Truth for the Test Set and Their Qualifications

Number of Experts: Three blinded radiologists.
Qualifications of Experts: Radiologists (no specific experience level mentioned, but "blinded radiologists" implies qualified professionals).

4. Adjudication Method for the Test Set

Adjudication Method: The ground truth was established by "three blinded radiologists." This implies a consensus method, likely majority rule or a process where discrepancies were resolved to arrive at a single ground truth label. The specific phrase "consensus" or "adjudication" is not used, but the description points to this approach.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study Was Done, and the Effect Size of How Much Human Readers Improve with AI vs Without AI Assistance

MRMC Study Done: Yes, a multi-reader multi-case study was conducted.
Effect Size of Human Reader Improvement with AI vs. Without AI Assistance:
- Mean AUC: Improved by 14.5% (from 76.8% non-aided to 91.3% aided; p=0.002).
- Sensitivity: Increased by 16.3% (from 67.4% non-aided to 83.7% aided; p<0.001).
- Specificity: Increased by 12.4% (from 76.6% non-aided to 89.0% aided; p<001).
- Large Pneumothorax (Mean AUC): Improved by 10.5% (p=0.009).
- Large Pneumothorax (Sensitivity): Improved by 13.4% (p<0.001).
- Small Pneumothorax (Mean AUC): Improved by 17.6% (p<0.001).
- Small Pneumothorax (Sensitivity): Improved by 18.7% (p<0.001).

6. If a Standalone (i.e., algorithm only without human-in-the-loop performance) Was Done

Standalone Study Done: Yes, the "standalone performance of the Pneumothorax Detection AI Algorithm was tested against this dataset."

7. The Type of Ground Truth Used

Type of Ground Truth: Expert consensus by three blinded radiologists.

8. The Sample Size for the Training Set

Sample Size for Training Set: The algorithm was developed using "over 12,000 images." This number includes images used for training, verification, and validation, but the specific breakdown for the training set alone is not provided. It's implied that the majority would be for training.

9. How the Ground Truth for the Training Set Was Established

Ground Truth for Training Set: The document states that the "Pneumothorax Detection AI Algorithm was developed using over 12,000 images from six sources, including the National Institute of Health and sites within the United States, Canada, and India." It then clarifies this data was "segregated into training, verification, and validation datasets." While it doesn't explicitly detail the methodology for establishing ground truth for the training set, it's standard practice that such large datasets for deep learning and medical imaging are meticulously annotated by medical experts (e.g., radiologists) or derived from existing clinical reports and pathology, which would then be reviewed or confirmed by experts. Given the rigor for the validation set, it's reasonable to infer a similar expert-driven process for the training data, although the specifics are not provided in this excerpt.

Summary

{0}------------------------------------------------

Image /page/0/Picture/0 description: The image contains the logo of the U.S. Food & Drug Administration (FDA). On the left is the Department of Health & Human Services logo. To the right of that is the FDA logo, which is a blue square with the letters "FDA" in white. To the right of the blue square is the text "U.S. FOOD & DRUG ADMINISTRATION" in blue.

GE Medical Systems, LLC % Chris Paulik Regulatory Affairs Manager 3000 N. Grandview Blvd. WAUKESHA WI 53188

May 25, 2023

Re: K223491

Trade/Device Name: Critical Care Suite with Pneumothorax Detection AI Algorithm Regulation Number: 21 CFR 892.2090 Regulation Name: Radiological computer assisted detection and diagnosis software Regulatory Class: Class II Product Code: QBS Dated: April 26, 2023 Received: April 27, 2023

Dear Chris Paulik:

We have reviewed your Section 510(k) premarket notification of intent to market the device referenced above and have determined the device is substantially equivalent (for the indications for use stated in the enclosure) to legally marketed predicate devices marketed in interstate commerce prior to May 28, 1976, the enactment date of the Medical Device Amendments, or to devices that have been reclassified in accordance with the provisions of the Federal Food, Drug, and Cosmetic Act (Act) that do not require approval of a premarket approval application (PMA). You may, therefore, market the device, subject to the general controls provisions of the Act. Although this letter refers to your product as a device, please be aware that some cleared products may instead be combination products. The 510(k) Premarket Notification Database located at https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm identifies combination product submissions. The general controls provisions of the Act include requirements for annual registration, listing of devices, good manufacturing practice, labeling, and prohibitions against misbranding and adulteration. Please note: CDRH does not evaluate information related to contract liability warranties. We remind you, however, that device labeling must be truthful and not misleading.

If your device is classified (see above) into either class II (Special Controls) or class III (PMA), it may be subject to additional controls. Existing major regulations affecting your device can be found in the Code of Federal Regulations, Title 21, Parts 800 to 898. In addition, FDA may publish further announcements concerning your device in the Federal Register.

Please be advised that FDA's issuance of a substantial equivalence determination does not mean that FDA has made a determination that your device complies with other requirements of the Act or any Federal statutes and regulations administered by other Federal agencies. You must comply with all the Act's requirements, including, but not limited to: registration and listing (21 CFR Part 807); labeling (21 CFR Part 801); medical device reporting of medical device-related adverse events) (21 CFR 803) for

{1}------------------------------------------------

devices or postmarketing safety reporting (21 CFR 4, Subpart B) for combination products (see https://www.fda.gov/combination-products/guidance-regulatory-information/postmarketing-safety-reportingcombination-products); good manufacturing practice requirements as set forth in the quality systems (QS) regulation (21 CFR Part 820) for devices or current good manufacturing practices (21 CFR 4, Subpart A) for combination products; and, if applicable, the electronic product radiation control provisions (Sections 531-542 of the Act); 21 CFR 1000-1050.

Also, please note the regulation entitled, "Misbranding by reference to premarket notification" (21 CFR Part 807.97). For questions regarding the reporting of adverse events under the MDR regulation (21 CFR Part 803), please go to https://www.fda.gov/medical-device-safety/medical-device-reportingmdr-how-report-medical-device-problems.

For comprehensive regulatory information about mediation-emitting products, including information about labeling regulations, please see Device Advice (https://www.fda.gov/medicaldevices/device-advice-comprehensive-regulatory-assistance) and CDRH Learn (https://www.fda.gov/training-and-continuing-education/cdrh-learn). Additionally, you may contact the Division of Industry and Consumer Education (DICE) to ask a question about a specific regulatory topic. See the DICE website (https://www.fda.gov/medical-device-advice-comprehensive-regulatoryassistance/contact-us-division-industry-and-consumer-education-dice) for more information or contact DICE by email (DICE@fda.hhs.gov) or phone (1-800-638-2041 or 301-796-7100).

Sincerely.

Jessica Lamb

Jessica Lamb, Ph.D. Assistant Director Imaging Software Team DHT8B: Division of Radiological Imaging Devices and Electronic Products OHT8: Office of Radiological Health Office of Product Evaluation and Quality Center for Devices and Radiological Health

Enclosure

{2}------------------------------------------------

Indications for Use

510(k) Number (if known)

223491

Device Name

Critical Care Suite with Pneumothorax Detection AI Algorithm

Indications for Use (Describe)

Intended users include qualified independently licensed healthcare professionals (HCPs) trained to independently assess the presence of pneumothoraxes in radiographic images and radiologists.

Type of Use (Select one or both, as applicable)
☑ Prescription Use (Part 21 CFR 801 Subpart D) ☐ Over-The-Counter Use (21 CFR 801 Subpart C)

CONTINUE ON A SEPARATE PAGE IF NEEDED.

{3}------------------------------------------------

Image /page/3/Picture/0 description: The image shows the General Electric (GE) logo. The logo consists of the letters 'GE' in a stylized script, enclosed within a blue circle. There are also several droplet-like shapes surrounding the circle, giving it a dynamic and fluid appearance. The logo is simple, recognizable, and associated with a well-known multinational corporation.

K223491

510(k) Summary

In accordance with 21 CFR 807.92 the following summary of information is provided:

Date:	May 25, 2023
Submitter:	GE HealthCare, (GE Medical Systems, LLC)
	3000 N. Grandview Blvd
	Waukesha, WI 53188 USA
PrimaryContactPerson:	Chris PaulikSenior Regulatory Affairs ManagerGE HealthCare262-894-5415Christopher.A.Paulik@ge.com
SecondaryContactPerson:	Gregory PessatoRegulatory Affairs DirectorGE HealthCare+33 (6) 34423240GregoryPessato@ge.com
Device TradeName:	Critical Care Suite with Pneumothorax Detection Al Algorithm
Common /Usual Name:	Radiological computer assisted detection and diagnosis software
ClassificationNames andProduct Code:	Regulation Name: Radiological computer assisted detection and diagnosis softwareRegulation: 21 CFR 892.2090Classification: Class IIProduct Codes: QBS
PredicateDevice:	BoneView (K212365)Regulation Name: Radiological computer assisted detection and diagnosis softwareRegulation: 21 CFR 892.2090Classification: Class II
	Product Codes: QBS
ReferenceDevice:	Critical Care Suite (K183182)Regulation Name: Radiological computer aided triage and notification softwareRegulation: 21 CFR 892.2080Classification: Class IIProduct Codes: QFM
DeviceDescription:	Critical Care Suite is a suite of Al algorithms for the automated image analysis of frontal chest X-rays acquired on a digital x-ray system for the presence of critical findings. Critical Care Suite with Pneumothorax Detection Al Algorithm is indicated for adults and transitional adolescents (18 to <22 years old but treated like adults) and is intended to be used by licensed qualified healthcare professionals (HCPs) trained to independently assess the presence of pneumothoraxes in radiographic images and radiologists. Critical Care Suite is a software module that can be deployed on several computing platforms such as PACS, On Premise, On Cloud or X-ray Imaging Systems.Today's clinical workflow, hospitals are overburdened by large volume of orders and long turnaround times for radiologist reports. Critical Care Suite with the Pneumothorax Detection Al Algorithm enables effective prioritization and assists in the detection / diagnosis of pneumothoraxes for radiologists and HCPs that have been trained to independently assess the presence of pneumothoraxes in radiographic images. It performs this task by flagging images with a suspicious finding and providing a localization overlay of the suspected pneumothorax as well as a graphical representation of the algorithm's confidence in the resultant finding. These outputs can be displayed wherever the reviewing physician normally conducts their reads per their standard of care, including PACS, On Premise, On Cloud and Digital Projection Radiographic Systems.
Intended Use:	Critical Care Suite with Pneumothorax Detection Al Algorithm is intended to aide a clinician in the detection and localization of a pneumothorax on frontal chest radiographic images.
Indications forUse:	Critical Care Suite with Pneumothorax Detection Al Algorithm is a computer-aided triage, notification, and diagnostic device that analyzes frontal chest X-ray images for the presence of a pneumothorax. Critical Care Suite identifies and highlights images with a pneumothorax to enable case prioritization or triage and assist as a concurrent reading aid during interpretation of radiographs.Intended users include qualified independently licensed healthcare professionals (HCPs) trained to independently assess the presence of pneumothoraxes in radiographic images and radiologists.Critical Care Suite should not be used in-lieu of full patient evaluation or solely relied upon to make or confirm a diagnosis. It is not intended to replace the review of the X-ray
	image by a qualified physician. Critical Care Suite is indicated for adults and TransitionalAdolescents (18 to <22 years old but treated like adults).
Technology:	Critical Care Suite with Pneumothorax Detection Al Algorithm employs the samefundamental scientific technology as its predicate and reference devices. They are alldeep learning locked AI algorithms that can be deployed on several computing platformssuch as PACS, On Premise, On Cloud or X-ray Imaging Systems. The patient and userpopulations are equivalent to what was provided with Critical Care Suite withPneumothorax Detection Al Algorithm. The output is equivalent since both predicateand proposed devices produce a result if a suspicious finding is detected, provide alocalization overlay of the suspected pathology within the image and a representation ofthe algorithm's confidence in the resultant finding. The intended use has been expandedfrom the original release of Critical Care Suite (K183182) to display an overlay to thereviewing physician that helps localize a detected pneumothorax. It also provides aconfidence level to the reviewing physician that provides contextual information in thealgorithm's confidence for its pneumothorax detection output.
	The differences between Critical Care Suite with Pneumothorax Detection Al Algorithmand BoneView are the specific pathologies that are being detected. Critical Care Suitewith Pneumothorax Detection Al Algorithm analyzes frontal chest radiographic imagesfor the presence of a suspected pneumothorax where BoneView analyzes radiographicimages for the presence of suspected fractures. This difference does not impact thesafety or efficacy of Critical Care Suite with Pneumothorax Detection Al Algorithm sinceboth devices analyze images using deep learning Al technology to detect pathologiesproducing an output that can aide clinicians and radiologists with their diagnosis.

{4}------------------------------------------------

Image /page/4/Picture/0 description: The image shows the General Electric (GE) logo. The logo consists of the letters 'GE' written in a stylized, cursive font, enclosed within a blue circle. The circle is surrounded by a series of swirling, water-like shapes, also in blue, that give the impression of movement or energy emanating from the central letters.

GE HealthCare 510(k) Premarket Notification Submission

{5}------------------------------------------------

Image /page/5/Picture/1 description: The image shows the logo for General Electric (GE). The logo is a blue circle with the letters "GE" in a stylized font in the center. There are three water droplet shapes surrounding the circle. The logo is simple and recognizable, and it is often used to represent the company's brand.

510(k) Premarket Notification Submission

Product DeviceComparison	Critical Care Suite with PneumothoraxDetection AI Algorithm	BoneView (K212365)
DeviceClassification	Radiological computer assisted detection anddiagnosis softwareClass II, QBS	Radiological computer assisted detection anddiagnosis softwareClass II, QBS
Targeted clinicalcondition,anatomy, andimaging modality	PneumothoraxChest/LungAP/PA Chest X-Ray Imaging	FractureAnkle, Foot, Knee, Femur, Wrist, Hand, Elbow,Forearm, Humerus, Shoulder, Clavicle, Pelvis, Hip,Ribs, Thoracic Spine, Lumbar Spine2D Radiographic Images
AlgorithmInferencingMechanism	AI deep learning algorithms designed to detectpneumothorax in frontal chest X-ray images to aidein identifying and highlighting pneumothoraxesduring the review of radiographs.	Al supervised deep learning algorithm designed toaide in identifying and highlighting fractures duringthe review of radiographs.
ComputationalPlatform	Critical Care Suite is designed as a self-containedsoftware module deployable on variouscomputational and x-ray imaging system platforms	Deployment on-premises or on cloud and connectionto several computing platforms and X-ray imaging

{6}------------------------------------------------

Image /page/6/Picture/0 description: The image shows the logo for General Electric (GE). The logo is a blue circle with the letters "GE" in a stylized font in the center. There are three white water droplets surrounding the letters. The logo is simple and recognizable, and it is associated with a well-known company.

GE HealthCare 510(k) Premarket Notification Submission

Product DeviceComparison	Critical Care Suite with PneumothoraxDetection AI Algorithm	BoneView (K212365)
	such as Digital Projection Radiographic Systems,PACS, On Premise or On Cloud.	platforms such as X-ray radiographic systems, orPACS
Algorithm Outputs	1. Configurable DICOM tag that identifies if asuspected pneumothorax was detected.2. Image annotations that contain:Flag if a suspected pneumothorax wasdetected Graphical representation of thealgorithms confidence in the algorithmsresult Overlay (color or grayscale) that localizesthe pneumothorax within the image	1. Optional Summary Table with the results of theoverall study2. Results Image that containsRegion of Interest that is a solid or dottedrectangle based on confidence of thealgorithm Summary including the number of regionsof interest that are displayed and a cautionmessage if it was identified that the imagewas not part of the indications for use ofBoneView.
Destination forViewing AlgorithmResults	Image annotation on a secondary DICOM image anda DICOM message that identifies if a suspectedpneumothorax was detected within the study.The output can be immediately used to visualize theresults on any DICOM destination such as a user'simages storage system (PACS) or the x-ray system.	Image annotations made on copy of original image orimage annotations toggled on/off.The output can be immediately used to visualize theresult on any DICOM destination such as a user'simages storage system (PACS) or other radiologicalequipment (X-Ray System)

Clinical andNon-ClinicalTests:	Summary of Non-Clinical Tests:
	The following quality assurance measures were applied to the development of CriticalCare Suite with Pneumothorax Detection AI Algorithm and deployment onto the AMXNavigate system:
	1. Risk Analysis
	2. Requirements Reviews
	3. Design Reviews
	4. Testing on unit level (Module verification)
	5. Integration testing (System verification)
	6. Performance testing (Verification)
	7. Safety testing (Verification)
	8. Simulated use testing (Validation)
	Critical Care Suite with Pneumothorax Detection AI Algorithm specific verification wasconducted to demonstrate proper implementation of Critical Care Suite software designrequirements.

{7}------------------------------------------------

Image /page/7/Picture/1 description: The image shows the logo for General Electric (GE). The logo consists of the letters "GE" written in a stylized, cursive font. The letters are enclosed within a circle, and there are three water droplet-like shapes surrounding the circle. The logo is blue.

Regression testing on the AMX Navigate feature functionality was conducted to verify proper integration of Critical Care Suite with Pneumothorax Detection Al Algorithm into the AMX Navigate software and device. Validation was performed on AMX Navigate with integrated Critical Care Suite with Pneumothorax Detection Al Algorithm. Design verification and validation testing was performed to confirm that the safety and effectiveness of the device has not been affected. The test plans and results have been executed with acceptable results. Summary of Clinical Tests: The Pneumothorax Detection Al Algorithm was developed using over 12,000 images from six sources, including the National Institute of Health and sites within the United States, Canada, and India. This data was then segregated into training, verification, and validation datasets. The final validation ground truth dataset included 804 images from two North American sites that were not used in the training process of the algorithm. A mix of cases with low, moderate, and high complexity were included in the dataset. 544 images were acquired on GE HealthCare scanners and 264 images acquired on non-GE Healthcare scanners. Only one site was able to provide age and gender demographics which included a distribution of 51.2% males and 48.8% females, with a median age of 68 (min 18, max 90+). The reference standard was established by three blinded radiologists. The standalone performance of the Pneumothorax Detection Al Algorithm was tested against this dataset establishing that the algorithm can detect a pneumothorax within a frontal chest x-ray image and that the Pneumothorax Overlay can localize a suspected pneumothorax. The ground truth dataset adequately analyzed all the primary and secondary endpoints and the results met the defined passing criteria. The Pneumothorax Detection Al Algorithm achieved an AUC of 96.1% (94.9%, 97.2%), a sensitivity of 84.3% (80.6%, 88.0%) and a specificity of 93.2% (90.8%, 95.6%) for detection of pneumothoraxes on both anteroposterior and posteroanterior frontal chest x-ray images. The algorithm also had high sensitivity for detecting large pneumothoraxes with a sensitivity of 96.3% (93.1%, 99.2%) and small pneumothorax with a sensitivity of 75.0% (69.2%, 80.8%). Additionally, the Pneumothorax Overlay was assessed on the true positive cases identified above, and it partially localized 98.1% (96.6%, 99.6%) of the actual pneumothorax within an image between the apical, lateral, and inferior regions of a lung. It performed with full agreement between these regions 67.8% (62.7%, 73.0%). lt also performed with a DICE Similarity Coefficient of 0.705 (0.683, 0.724) indicating that on average 70.5% of the Pneumothorax Overlay area and the true area of a pneumothorax within an image overlap. A multi-reader multi-case study was conducted to assess that the use of the Critical Care Suite with Pneumothorax Detection Al Algorithm improves reader performance within the intended use population in detecting / diagnosing a pneumothorax in a frontal chest x-ray image. This study consisted of 10 independent readers to adequately analyze all the primary and secondary endpoints of varied experiences levels representing the

{8}------------------------------------------------

Image /page/8/Picture/1 description: The image shows the logo for General Electric (GE). The logo is a blue circle with the letters "GE" in a stylized font in the center. There are also some white swirls around the letters, which give the impression of movement or energy. The logo is simple and recognizable, and it is often used to represent the company's products and services.

	clinical users who would interact with the Critical Care Suite with PneumothoraxDetection Al Algorithm: radiologists (Rad.), internal medicine (IM) physicians, emergencymedicine (ER) physicians, and nurse practitioners. This study contained 400 images fromthe original validation ground truth dataset used to determine the standaloneperformance of the algorithm, and adequately analyzed that all the primary andsecondary endpoints met the defined passing criteria.
	Critical Care Suite with Pneumothorax Detection Al Algorithm improved readerperformance for detection of pneumothorax, measured by mean AUC, by 14.5%(7.0%,22.0%; p=.002), from 76.8% non-aided to 91.3% aided. Reader sensitivityincreased by 16.3% (13.1%, 19.5%; p<.001) from 67.4% non-aided to 83.7% aided.Reader specificity increased by 12.4% (9.6%, 15.1%; p<.001) from 76.6% non-aided to89.0% aided. The overall performance by size was also improved. The readers showedan improvement for detection of large pneumothorax measured by mean AUC 10.5%(3.2%, 17.8%, p=0.009) and sensitivity 13.4% (10.0%, 16.9%, p<.001). The readersshowed an improvement for detection of small pneumothorax measured by mean AUC17.6% (9.3%, 25.9%, p<0.001) and sensitivity 18.7% (13.8%, 23.6%, p<.001). Thedifferent clinical user's improvements in mean AUC were assessed, and it was noted thatall physicians (Rad, IM, ER) improved 10.4% (2.8%, 17.9%, p=0.015), nurse practitionersimproved 24.1% (1.2%, 47.0%, p=0.045), radiologists improved 2.4% (-1.0%, 5.7%,p=0.095), and non-radiologists (ER, IM, NP) improved 17.5% (9.6%, 25.4%, p<0.001).
Determinationof SubstantialEquivalence:	The introduction of Critical Care Suite with Pneumothorax Detection Al Algorithm doesnot result in any new potential safety risks and uses the same fundamental deep learningbased technology to detect pathological finding on 2D X-ray images. Technologicaldifferences were assessed through bench testing and clinical validation. Like its predicatethe device has been shown to improve intended user accuracy at detecting the targetedpathological finding by licensed healthcare professionals, thus demonstrating that theproposed device is substantially equivalent to its predicate.
	After analyzing design verification and validation testing on the bench and the clinicaltesting results it is the conclusion of GE HealthCare that the Critical Care Suite withPneumothorax Detection AI Algorithm software to be as safe, as effective, andperformance is substantially equivalent to the predicate device.

Regulation Number and Section

§ 892.2090 Radiological computer-assisted detection and diagnosis software.

(a)
Identification. A radiological computer-assisted detection and diagnostic software is an image processing device intended to aid in the detection, localization, and characterization of fracture, lesions, or other disease-specific findings on acquired medical images (e.g., radiography, magnetic resonance, computed tomography). The device detects, identifies, and characterizes findings based on features or information extracted from images, and provides information about the presence, location, and characteristics of the findings to the user. The analysis is intended to inform the primary diagnostic and patient management decisions that are made by the clinical user. The device is not intended as a replacement for a complete clinician's review or their clinical judgment that takes into account other relevant information from the image or patient history.(b)
Classification. Class II (special controls). The special controls for this device are:(1) Design verification and validation must include:
(i) A detailed description of the image analysis algorithm, including a description of the algorithm inputs and outputs, each major component or block, how the algorithm and output affects or relates to clinical practice or patient care, and any algorithm limitations.
(ii) A detailed description of pre-specified performance testing protocols and dataset(s) used to assess whether the device will provide improved assisted-read detection and diagnostic performance as intended in the indicated user population(s), and to characterize the standalone device performance for labeling. Performance testing includes standalone test(s), side-by-side comparison(s), and/or a reader study, as applicable.
(iii) Results from standalone performance testing used to characterize the independent performance of the device separate from aided user performance. The performance assessment must be based on appropriate diagnostic accuracy measures (
e.g., receiver operator characteristic plot, sensitivity, specificity, positive and negative predictive values, and diagnostic likelihood ratio). Devices with localization output must include localization accuracy testing as a component of standalone testing. The test dataset must be representative of the typical patient population with enrichment made only to ensure that the test dataset contains a sufficient number of cases from important cohorts (e.g., subsets defined by clinically relevant confounders, effect modifiers, concomitant disease, and subsets defined by image acquisition characteristics) such that the performance estimates and confidence intervals of the device for these individual subsets can be characterized for the intended use population and imaging equipment.(iv) Results from performance testing that demonstrate that the device provides improved assisted-read detection and/or diagnostic performance as intended in the indicated user population(s) when used in accordance with the instructions for use. The reader population must be comprised of the intended user population in terms of clinical training, certification, and years of experience. The performance assessment must be based on appropriate diagnostic accuracy measures (
e.g., receiver operator characteristic plot, sensitivity, specificity, positive and negative predictive values, and diagnostic likelihood ratio). Test datasets must meet the requirements described in paragraph (b)(1)(iii) of this section.(v) Appropriate software documentation, including device hazard analysis, software requirements specification document, software design specification document, traceability analysis, system level test protocol, pass/fail criteria, testing results, and cybersecurity measures.
(2) Labeling must include the following:
(i) A detailed description of the patient population for which the device is indicated for use.
(ii) A detailed description of the device instructions for use, including the intended reading protocol and how the user should interpret the device output.
(iii) A detailed description of the intended user, and any user training materials or programs that address appropriate reading protocols for the device, to ensure that the end user is fully aware of how to interpret and apply the device output.
(iv) A detailed description of the device inputs and outputs.
(v) A detailed description of compatible imaging hardware and imaging protocols.
(vi) Warnings, precautions, and limitations must include situations in which the device may fail or may not operate at its expected performance level (
e.g., poor image quality or for certain subpopulations), as applicable.(vii) A detailed summary of the performance testing, including test methods, dataset characteristics, results, and a summary of sub-analyses on case distributions stratified by relevant confounders, such as anatomical characteristics, patient demographics and medical history, user experience, and imaging equipment.