Predicate Devices

Reference Devices

Not Found

AI/ML (Artificial Intelligence / Machine Learning)

Yes
The summary explicitly states that the device utilizes "deep learning technology" and "machine learning (more specifically, supervised deep learning) methodology" for image processing and abnormality detection.

Therapeutic

No
The device aids in the detection and diagnosis of abnormalities (ARIA) from MR images, but it does not directly treat or prevent a disease, nor does it restore, modify, or correct body function or structure. It is a diagnostic aid, not a therapeutic device.

Diagnostic

Yes.
The "Intended Use / Indications for Use" section explicitly states that the device is for "computer-assisted detection (CADe) and diagnosis (CADx) software device". It also mentions providing information about the "presence, location, size, severity and changes" of certain abnormalities, which are all characteristics of a diagnostic device.

SaMD (Software as a Medical Device)

Yes

The device description explicitly states "icobrain aria is a software-only device". The rest of the summary details its software functions, inputs (DICOM images), outputs (annotated images and reports), and performance studies, without mentioning any associated hardware components that are part of the medical device itself.

IVD (In Vitro Diagnostic)

Based on the provided information, this device is not an In Vitro Diagnostic (IVD).

Here's why:

IVDs analyze biological samples: In Vitro Diagnostics are designed to examine specimens taken from the human body, such as blood, urine, tissue, etc., to provide information about a person's health.
This device analyzes medical images: icobrain aria processes brain MRI images, which are medical images, not biological samples. It analyzes the visual information within the images to detect and characterize abnormalities.

The device falls under the category of medical image analysis software or computer-assisted detection/diagnosis software, which are distinct from IVDs.

PCCP (Predetermined Change Control Plan) Authorized

No
The letter does not explicitly state that the FDA has reviewed and approved or cleared a Predetermined Change Control Plan (PCCP) for this specific device. While "PCCP" is a keyword in the prompt, there is no mention of the device being authorized as a PCCP.

Overview

Intended Use / Indications for Use

icobrain aria is a computer-assisted detection (CADe) and diagnosis (CADx) software device to be used as a concurrent reading aid to help trained radiologists in the detection, assessment and characterization of Amyloid Related Imaging Abnormalities (ARIA) from a set of brain MR images. The software provides information about the presence, location, size, severity and changes of ARIA-E (brain edema or sulcal effusions) and ARIA-H (hemosiderin deposition, including microhemorrhage and superficial siderosis). Patient management decisions should not be made solely on the basis of analysis by icobrain aria.

Product codes (comma separated list FDA assigned to the subject device)

QBS

Device Description

icobrain aria is a software-only device for assisting radiologists with the detection of amyloid-related imaging abnormalities (ARIA) on brain MRI scans of Alzheimer's disease patients under an amyloid beta-directed antibody therapy. The device utilizes 2D fluid-attenuated inversion recovery (FLAR) for the detection of ARIA-E (edema/sulcal effusion) and 2D T2* gradient echo (T2*-GRE) for the detection of ARIA-H (hemosiderin deposition).

icobrain aria automatically processes input brain MRI scans in DICOM format from two time points and generates annotated DICOM images and an electronic report. The following main measurements are included in the icobrain aria electronic report:

ARIA-E: the length of the longest axis computed from the segmented ARIA-E abnormalities, and the number of brain sites affected by ARIA-E;
ARIA-H: the count of stable and new T2*-GRE hypointensities indicated as microhemorrhages or superficial siderosis.

From these measurements, ARIA radiographic severities for ARIA-E, ARIA-H microhemorrhages and ARIA-H superficial siderosis are automatically derived and included in the icobrain aria electronic report. The ARA radiographic severity definitions implement the suggested categorization that has been incorporated in U.S. prescribing information for currently FDA-approved amyloid betadirected antibody therapies, such as Aduhelm® and also published more generally (Cogswell et al, 2022; doi: 10.3174/ ainr.A7586).

Mentions image processing

Yes

Mentions AI, DNN, or ML

Yes - "The image processing implemented in icobrain aria for ARIA-E and ARIA-H detection and segmentation is based on deep learning technology." and "both devices employ machine learning (more specifically, supervised deep learning) methodology for detecting and diagnosing abnormalities in medical images."

Input Imaging Modality

MR images; FLAIR; T2*-GRE

Anatomical Site

Brain

Indicated Patient Age Range

The training set included subjects aged 51 to 85 for FLAIR images and 51 to 86 for T2*-GRE images. The test set mean (SD) age was 70.4 (7.2) years, ranging from 52 to 89.

Intended User / Care Setting

Trained radiologists / Not specified

Description of the training set, sample size, data source, and annotation protocol

Representative data was used for developing the algorithms and consisted of brain MRI scans of patients from the aducanumab clinical trials PRIME (NCT02677572), EMERGE (NCT02484547) and ENGAGE (NCT02477800), which have been manually annotated by expert neuroradiologists with prior experience of reading ARIA in clinical trials of amyloid beta-directed antibody drugs. In total, 475 FLAIR image pairs from 172 subjects aged 51 to 85, and 326 T2*-GRE image pairs from 177 subjects aged 51 to 86, were used for the model development. Among the selected data for training, 81 cases were ARIA-free cases, for which 76.5% had FLAIR hyperintensities that were not labeled as ARIA by the experts.

Description of the test set, sample size, data source, and annotation protocol

Data characteristics: The study data population consists of MRI datasets from subjects diagnosed with Alzheimer's disease. To guarantee independence of test data from training data, MR images used for testing were acquired from subjects that were not included in the training set. MRI scans originated from more than 100 sites in 20 countries, with approximately half the data originating from the US and the other half from outside the US. The protocols used to collect images match the required acquisition parameters of the finished software as intended for clinical use (cfr. HCP manual).

The main test set on which standalone performance results are reported coincides with the testing (see section below, where a direct comparison is made between performance of radiologists assisted by the software, and the standalone software itself). For testing the ARIA-H models, the dataset consisted of 199 pairs of MRI scans from 199 individual subjects. Mean (SD) age was 70.4 (7.2) years, ranging from 52 to 89; 105 (52.8%) were Asian, 1 (0.5%) was Black, 157 (78.9%) were White, and 18 (9.0%) were other or unreported race and ethnicity. The scans were acquired in 33 distinct MRI scanners manufactured by GE (26%), Philips (17%), at 1.5 Tesla (44%) and 3 Tesla (56%).

An extended test set was also available for additional standalone performance testing. This larger dataset had similar demographics characteristics as the main test set: mean (SD) age was 70.6 (7.57), ranging from 52 to 89; 53% female; 75% White, 12% Asian, 1% Black or African-American, and 12% not reported. The scans were acquired in MRI scanners manufactured by GE (24%), Philips (19%) and Siemens (57%), at 1.5 Tesla (42%) and 3 Tesla (58%). For testing the ARIA-E models, the dataset consisted of 1015 pairs of MRI scans from 385 individual subjects. For testing the ARA-H models, the dataset consisted of 449 pairs of MRI scans from 369 individual subjects.

Methods: The standalone testing consisted of comparing software outputs against expert manual annotations. To establish the ground truth, expert neuroradiologists (with experience performing safety ARIA reading in clinical trials for Aβ-directed antibody therapies in AD) manually segmented both ARIA-E and ARIA-H findings. Ground truth ARIA measurements were derived from the expert manual annotated masks.

Summary of Performance Studies (study type, sample size, AUC, MRMC, standalone performance, key results)

Standalone Performance Testing:
Study Type: Standalone performance assessments comparing software outputs against expert manual annotations.
Sample Size: 199 pairs of MRI scans from 199 individual subjects for ARIA-H models, and 199 cases (123 positive) for ARIA-E diagnostic performance. An extended test set included 1015 pairs of MRI scans from 385 subjects for ARIA-E and 449 pairs of MRI scans from 369 subjects for ARIA-H.
AUC:
ARIA-E: sensitivity of 0.94 (95% Cl [0.90, 0.98]), specificity 0.67 (95% Cl [0.56, 0.77]), and AUC 0.84 (95% Cl [0.77, 0.89]).
ARIA-H (pooled): sensitivity of 0.87 (95% Cl [0.80, 0.93]), specificity 0.66 (95% Cl [0.55, 0.76]), and AUC 0.81 (95% Cl [0.75, 0.86]).
ARIA-H microhemorrhages: sensitivity of 0.89 (95% Cl [0.83, 0.92]), specificity of 0.62 (95% Cl [0.54, 0.67]), and AUC of 0.80 (95% Cl [0.74, 0.85]).
ARIA-H superficial siderosis: sensitivity of 0.67 (95% Cl [0.57, 0.76]), specificity of 0.95 (95% Cl [0.91, 0.98]), and AUC of 0.82 (95% Cl [0.76, 0.87]).
Key Results: The standalone performance testing results demonstrated that the software detects ARIA-E and ARIA-H at case level and individual finding level in line with the performance of human experts.

Clinical Testing (MRMC Study):
Study Type: Fully-crossed multiple-case (MRMC) retrospective reader study.
Sample Size: 199 cases, 16 U.S. Board of Radiologists.
MRMC: Conducted as a multiple-reader multiple-case reader study.
AUC:
ARIA-E detection: assisted AUC 0.873 (95% Cl [0.835, 0.911]), AUC difference of 0.051 (95% Cl [0.020, 0.083]) compared to unassisted (p=0.001).
ARIA-H (pooled) detection: assisted AUC 0.825 (95% C [0.781, 0.869]), AUC difference of 0.044 (95% Cl [0.017, 0.070]) compared to unassisted (p=0.001).
ARIA-H microhemorrhages: assisted AUC 0.808 (95% Cl [0.760, 0.855]), AUC difference of 0.029 (95% Cl [0.002, 0.055]) compared to unassisted (p=0.032).
ARIA-H superficial siderosis: assisted AUC 0.784 (95% Cl [0.732, 0.836]), AUC difference of 0.063 (95% Cl [0.023, 0.102]) compared to unassisted (p=0.003).
Key Results: Radiologists assisted by icobrain aria were significantly better in detecting ARIA. Sensitivity increased significantly for all ARIA types, while specificity remained above 80% for most detections. Inter-reader variability was significantly lower for assisted reads, and readers were on average faster.

Key Metrics (Sensitivity, Specificity, PPV, NPV, etc.)

Standalone Performance:
ARIA-E: sensitivity 0.94, specificity 0.67
ARIA-H (pooled): sensitivity 0.87, specificity 0.66
ARIA-H microhemorrhages: sensitivity 0.89, specificity 0.62
ARIA-H superficial siderosis: sensitivity 0.67, specificity 0.95
ARIA-E finding-level true positive rate: 69.1% with 0.7 false positive findings per case.
ARIA-H new microhemorrhages finding-level true positive rate: 66.1% with 0.9 false positive findings per case.
ARIA-H new superficial siderosis finding-level true positive rate: 62.5% with 0.1 false positive findings per case.

Clinical Performance (Assisted vs. Unassisted):
ARIA-E detection: Sensitivity increased from 70.9% (unassisted) to 86.5% (assisted). Specificity remained above 80% (assisted: 0.830; unassisted: 0.917).
ARIA-H (pooled) detection: Sensitivity increased from 68.7% (unassisted) to 79.0% (assisted). Specificity remained above 80% (assisted: 0.803; unassisted: 0.828).
ARIA-H microhemorrhages detection: Sensitivity increased from 69.3% (unassisted) to 79.6% (assisted). Specificity was 76.7% (assisted) and 83.1% (unassisted).
ARIA-H superficial siderosis detection: Sensitivity increased from 49.7% (unassisted) to 59.9% (assisted). Specificity was 95.6% (assisted) and 92.7% (unassisted).
Inter-reader variability: Kendall's coefficient of concordance improved from 0.720 (unassisted) to 0.809 (assisted) for ARIA-E severity, and 0.656 (unassisted) to 0.799 (assisted) for ARIA-H severity.
Reading time: Median unassisted reading 2:34min; median assisted reading 2:21min.

Predicate Device(s): If the device was cleared using the 510(k) pathway, identify the Predicate Device(s) K/DEN number used to claim substantial equivalence and list them here in a comma separated list exactly as they appear in the text. List the primary predicate first in the list.

OsteoDetect DEN180005

Reference Device(s): Identify the Reference Device(s) K/DEN number and list them here in a comma separated list exactly as they appear in the text.

Not Found

Predetermined Change Control Plan (PCCP) - All Relevant Information for the subject device only (e.g. presence / absence, what scope was granted / cleared under the PCCP, any restrictions, etc).

Not Found

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.

Summary

0

Image /page/0/Picture/0 description: The image contains the logos of the Department of Health and Human Services and the Food and Drug Administration (FDA). The Department of Health and Human Services logo is on the left, and the FDA logo is on the right. The FDA logo includes the letters "FDA" in a blue square, followed by the words "U.S. Food & Drug Administration" in blue text.

November 7, 2024

icometrix NV Dirk Smeets CTO Kolonel Begaultlaan 1b/12 Leuven, Vlaams-Brabant 3012 Belgium

Re: K240712

Trade/Device Name: icobrain aria Regulation Number: 21 CFR 892.2090 Regulation Name: Radiological Computer Assisted Detection And Diagnosis Software Regulatory Class: Class II Product Code: QBS Dated: September 30, 2024 Received: September 30, 2024

Dear Dirk Smeets:

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 (the 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 available 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.

Additional information about changes that may require a new premarket notification are provided in the FDA guidance documents entitled "Deciding When to Submit a 510(k) for a Change to an Existing Device"

1

(https://www.fda.gov/media/99812/download) and "Deciding When to Submit a 510(k) for a Software Change to an Existing Device" (https://www.fda.gov/media/99785/download).

Your device is also subject to, among other requirements, the Quality System (QS) regulation (21 CFR Part 820), which includes, but is not limited to, 21 CFR 820.30. Design controls; 21 CFR 820.90. Nonconforming product; and 21 CFR 820.100, Corrective and preventive action. Please note that regardless of whether a change requires premarket review. the OS regulation requires device manufacturers to review and approve changes to device design and production (21 CFR 820.30 and 21 CFR 820.70) and document changes and approvals in the device master record (21 CFR 820.181).

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 Part 803) for devices or postmarketing safety reporting (21 CFR Part 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 Part 4, Subpart A) for combination products; and, if applicable, the electronic product radiation control provisions (Sections 531-542 of the Act); 21 CFR Parts 1000-1050.

All medical devices, including Class I and unclassified devices and combination product device constituent parts are required to be in compliance with the final Unique Device Identification System rule ("UDI Rue"). The UDI Rule requires, among other things, that a device bear a unique device identifier (UDI) on its label and package (21 CFR 801.20(a)) unless an exception or alternative applies (21 CFR 801.20(b)) and that the dates on the device label be formatted in accordance with 21 CFR 801.18. The UDI Rule (21 CFR 830.300(a) and 830.320(b)) also requires that certain information be submitted to the Global Unique Device Identification Database (GUDID) (21 CFR Part 830 Subpart E). For additional information on these requirements, please see the UDI System webpage at https://www.fda.gov/medical-device-advicecomprehensive-regulatory-assistance/unique-device-identification-system-udi-system.

Also, please note the regulation entitled, "Misbranding by reference to premarket notification" (21 CFR 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.

2

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,

Signature

Daniel M. Krainak, Ph.D. Assistant Director Magnetic Resonance and Nuclear Medicine Team DHT8C: Division of Radiological Imaging and Radiation Therapy Devices OHT8: Office of Radiological Health Office of Product Evaluation and Ouality Center for Devices and Radiological Health

Enclosure

3

Indications for Use

Submission Number (if known)

K240712

Device Name

icobrain aria

Indications for Use (Describe)

Intended use

icobrain aria is a computer-assisted detection (CADe) and diagnosis (CADx) software device to be used as a concurrent reading aid to help trained radiologists in the detection, assessment and characterization of Amyloid Related Imaging Abnormalities (ARIA) from a set of brain MR images. The software provides information about the presence, location, size, severity and changes of ARIA-E (brain edema or sulcal effusions) and ARIA-H (hemosiderin deposition, including microhemorrhage and superficial siderosis). Patient management decisions should not be made solely on the basis of analysis by icobrain aria.

Intended user population

The intended users of icobrain aria are trained radiologists. When a radiologist assesses the original MR images, both the icobrain aria annotated images and the original, unaltered images are used for concurrent reading.

Intended patient population

icobrain aria is intended to be used in the population of patients with Alzheimer's disease receiving treatment with an amyloid beta-directed antibody using the same ARIA radiographic severity scale as described in Aduhelm®/Legembi® prescription information, and presenting for safety monitoring, including severity assessment and monitoring changes of ARIA.

Warnings and precautions

icobrain aria is an adjunct tool and is not intended to replace a radiologist's review of the images or his or her clinical judgment. icobrain aria is not intended as an independent interpretation of MR images. Decisions should not be made solely based on analysis by icobrain aria.

The device is not intended to be used to segment macrohemorrhages with a diameter of 10 mm or more).

Type of Use (Select one or both, as applicable)

ription Use (Part 21 CFR 801 Subpart D)

e-Counter Use (21 CFR 801 Subpart C)

CONTINUE ON A SEPARATE PAGE IF NEEDED.

4

Image /page/4/Picture/0 description: The image shows the logo for icometrix. The logo consists of a stylized, abstract shape resembling a brain or a swirling pattern in shades of blue, green, and red. To the right of the shape is the word "icometrix" in a clean, sans-serif font, with all letters in lowercase and a bold, black color.

510(k) Summary

K240712

1 Submitter

Name	icometrix NV
Address:	Kolonel Begaultlaan 1b / 12
3012 Leuven
BELGIUM
Contact Person:	Dirk Smeets
Telephone number:	+32 16 369 000
Fax Number:	N.A.
E-mail:	dirk.smeets@icometrix.com
Date Prepared:	12 Mar 2024

2 Device

Device Trade Name:	icobrain aria
Regulation Number	21 CFR 892.2090
Regulation Name	Radiological Computer Assisted Detection And Diagnosis Software
Regulatory Class	Class II
Product Code:	QBS
Classification Panel:	Radiology

3 Predicate Device

Device	OsteoDetect
De Novo Number	DEN180005
Manufacturer	Imagen Technologies
Regulation Number	21 CFR 892.2090
Regulation Name	Radiological Computer Assisted Detection And Diagnosis Software
Regulatory Class	Class II
Product Code:	QBS

4 Device Description

icobrain aria is a software-only device for assisting radiologists with the detection of amyloid-related imaging abnormalities (ARIA) on brain MRI scans of Alzheimer's disease patients under an amyloid beta-directed antibody therapy. The device utilizes 2D fluid-attenuated inversion recovery (FLAR) for the detection of ARIA-E (edema/sulcal effusion) and 2D T2* gradient echo (T2*-GRE) for the detection of ARIA-H (hemosiderin deposition).

icobrain aria automatically processes input brain MRI scans in DICOM format from two time points and generates annotated DICOM images and an electronic report. The following main measurements are included in the icobrain aria electronic report:

5

Image /page/5/Picture/0 description: The image shows the icometrix logo. The logo consists of a stylized brain shape on the left and the word "icometrix" on the right. The brain shape is made up of three curved lines in different colors: blue, green, and red. The word "icometrix" is written in a sans-serif font, with the letters in black.

ARIA-E: the length of the longest axis computed from the segmented ARIA-E abnormalities, and the number of brain sites affected by ARIA-E;
ARIA-H: the count of stable and new T2*-GRE hypointensities indicated as microhemorrhages or superficial siderosis.

From these measurements, ARIA radiographic severities for ARIA-E, ARIA-H microhemorrhages and ARIA-H superficial siderosis are automatically derived and included in the icobrain aria electronic report. The ARA radiographic severity definitions implement the suggested categorization that has been incorporated in U.S. prescribing information for currently FDA-approved amyloid betadirected antibody therapies, such as Aduhelm® and also published more generally (Cogswell et al, 2022; doi: 10.3174/ ainr.A7586).

Table: ARIA Radiographic Severity scale.
Type	ARIA Radiographic Severity
	Mild	Moderate	Severe
ARIA-E	one location with longest
axis 10 cm
ARIA-H microhemorrhage	1 to 4 new incident
microhemorrhages	5 to 9 new incident
microhemorrhages	10 or more new incident
microhemorrhages
ARIA-H superficial
siderosis	1 new focal area of
superficial siderosis	2 new focal areas of superficial
siderosis	> 2 new focal areas of
superficial siderosis

The image processing implemented in icobrain aria for ARIA-E and ARIA-H detection and segmentation is based on deep learning technology. Hyperintensities in FLAIR brain suggestive of ARIA-E, and/or hypointensities in T2 -GRE brain scans, suggestive of ARIA-H, are segmented and the brain is parcellated into 10 regions (left and right frontal, parietal, temporal, occipital lobes, cerebellum and the rest of the brain) in order to provide location-specific ARIA presence. Representative data was used for developing the algorithms and consisted of brain MRI scans of patients from the aducanumab clinical trials PRIME (NCT02677572), EMERGE (NCT02484547) and ENGAGE (NCT02477800), which have been manually annotated by expert neuroradiologists with prior experience of reading ARIA in clinical trials of amyloid beta-directed antibody drugs. In total, 475 FLAR image pairs from 172 subjects aged 51 to 85, and 326 T2* -GRE image pairs from 177 subjects aged 51 to 86, were used for the model development. Among the selected data for training, 81 cases were ARIA-free cases, for which 76.5% had FLAIR hyperintensities that were not labeled as ARIA by the experts.

6

Image /page/6/Picture/0 description: The image shows the logo for icometrix. The logo consists of a stylized, abstract shape resembling a brain or a curved leaf in gradient colors of teal, green, and red. To the right of the shape is the word "icometrix" in a simple, sans-serif font, with the letters in black.

4.1 Outputs produced by the software

The software outputs are annotated DICOM images and an eletronic report summarizing the automated finding. These outputs are reviewed by radiologists concurrently alongside the raw input image review software. Finally, the radiologist summarizes all findings (according to his/her best interpretation after reviewing all images and software outputs) in a structured report, which is then forwarded to the PACS.

Annotated DICOM images

The following color-coded images are generated from the image processing results. These annotated images are made available as an output to the user in DICOM format.

1. ARIA-E overlay: original FLAIR image from the second time point overlaid with the ARIA-E segmentation in 3D. The stable and increasing regions are marked in different colors (Orange = New or enlarging FLAIR ARIA-E hyperintensities; Green = Stable FLAIR ARIA-E hyperintensities).
1. ARIA-H overlay: original T2*-GRE image from the second time point overlaid with the ARIA-H (microhemorrhages and superficial siderosis, in different colors) segmentation in 3D. The stable and new ARIA-H regions are marked in different colors (Red = New ARIA-H microhemorrhages; Orange = New ARIA-H superficial siderosis; Green = Stable ARIA-H microhemorrhages; Cyan = Stable ARIA-H superficial siderosis).

FLAIR NEW-ENLARGING ARIA-E

Orange = New or enlarging FLAIR hyperintensities suggestive of ARIA-E
Green = Stable FLAIR hyperintensities . suggestive of ARIA-E

T2 *- GRE ARIA-H

Red = New T2 *-GRE hypointensity suggestive of microhemorrhage .
Orange = New T2 *-GRE hypointensity suggestive of superficial siderosis .
Green = Stable T2 *-GRE hypointensity suggestive of microhemorrhage .
Cyan = Stable T2*-GRE hypointensity suggestive of superficial siderosis

Image /page/6/Picture/15 description: This is a medical scan of a brain. The scan is an axial view, meaning it is a cross-section of the brain. There are two highlighted regions on either side of the brain. The highlighted regions are yellow and green.

Image /page/6/Picture/16 description: The image shows two axial slices of a brain MRI. The left image shows the brain at a lower level, with the temporal lobes and brainstem visible. A small red dot is visible in the left temporal lobe. The right image shows the brain at a higher level, with the frontal and parietal lobes visible. A small yellow area is visible in the left frontal lobe.

7

Image /page/7/Picture/0 description: The image shows the logo for icometrix. The logo consists of a stylized brain shape in shades of blue, green, and red, followed by the word "icometrix" in black, sans-serif font. The brain shape is positioned to the left of the text.

Report

The icobrain aria report has 2 pages that correspond to the two time-point analysis of FLAIR images (for findings suggestive of ARA-E) and T2* GRE images (for findings suggestive of ARIA-H). Patient identification and scan dates are clearly printed in the headers.

The ARIA-E page includes a visual representation of brain slices from the FLAIR images at two time points, with the ARIA-E segmentation overlaid in color on the image of the second time point. If ARIA-E is detected, the slice corresponding the ARIA-E finding with the longest axis is also represented as a line segment on the FLAR image slice corresponding to the second time point.

| SAMPLE
ARIA-E

NAME
icobrain aria | DATE OF BIRTH
1952-01-01 | STUDY DATES
2018-02-06 - 2018-11-21 | icobrain aria
by icometrix

ID
ICO-ID |----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | This report is a summary of the automated findings. Please review the original images and annotated images by the software. | | 2018-02-06 | | | | LONGEST AXIS
99 mm | 2 | SUGGESTED SEVERITY*
Moderate | | Frontal Lobe | 0 | Parietal Lobe | Occipital Lobe | Temporal Lobe | | | Cerebellum | | Other | | | | Whole Brain | | *Suggested severity is derived from the automated findings
• None: No ARIA-E
• Mild: ARIA-E in 1 location axis 10 cm |
-----------------------------|----------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| | |
| 2018-11-21 | 2018-11-21 |
| IMAGING FINDINGS SUGGESTIVE OF ARIA-E | |
| SITES OF INVOLVEMENT
|
| Left (ml) | | Right (ml) |
| | 0 (+2) |
| 0 (+20) | | 0 |
| 0 (+4) | | 0 |
| 0 (+28) | | 0 |
| | Total (ml) |
| | 0 |
| | 0 |
| | |
| | 0 (+54) |

Volumes at Time Point 1
Volume change | + Increasing: > +9.2 ml | Decreasing: 23 mm overestimation of longest ARIA-E axis. 10% of cases had

27 mm underestimation of longest ARIA-E axis. 10% of cases had > 5 ml overestimation of
volume. 10% of cases had > 28 ml underestimation of volume.

+/- within measurement error: -2.1 - +9.2 ml |

8

Image /page/8/Picture/0 description: The image shows the logo for icometrix. The logo consists of a stylized brain shape on the left, with the word "icometrix" in black font to the right of the brain shape. The brain shape is colored with a gradient, transitioning from blue at the top to green in the middle and red at the bottom.

The ARIA-H page includes a visual representation of brain slices from the T2*-GRE images at two time points, with the ARIA-H segmentation overlaid in colors on the image of the second time point. If ARIA-H is detected, the slice corresponding to the highest proportion of ARIA-H findings (microhemorrhages and/or superficial siderosis).

Image /page/8/Figure/2 description: This image is a medical report from icobrain aria, detailing imaging findings suggestive of ARIA-H, specifically Microhemorrhages and Superficial Siderosis. For Microhemorrhages, the new count is 6, with a suggested severity of Moderate, and for Superficial Siderosis, the new count is 0, with a suggested severity of None. The report also includes a breakdown of counts in different lobes of the brain, such as the Frontal, Parietal, Occipital, and Temporal Lobes, as well as the Cerebellum and other areas.

In case any of the input images is missing or does not satisfy the required acquisition protocol, the report contains a warning stating that a required image is missing, followed by a list of image acquisition requirements for that type of image. The other page, if input images are correct, is still provided.

9

Image /page/9/Picture/0 description: The image shows the logo for icometrix. The logo consists of a stylized, abstract shape on the left and the word "icometrix" on the right. The shape is a curved, semi-circular form with a gradient of colors, transitioning from blue at the top to green in the middle and red at the bottom. The word "icometrix" is written in a clean, sans-serif font, with all letters in lowercase and black.

5 Intended use

Intended use

icobrain aria is a computer-assisted detection (CADe) and diagnosis (CADx) software device to be used as a concurrent reading aid to help trained radiologists in the detection, assessment and characterization of Amyloid Related Imaging Abnormalities (ARMA) from a set of brain MR images. The software provides information about the presence, location, size, severity and changes of ARIA-E (brain edema or sulcal effusions) and ARIA-H (hemosiderin deposition, including microhemorrhage and superficial siderosis). Patient management decisions should not be made solely on the basis of analysis by icobrain aria.

Intended user population

The intended users of icobrain aria are trained radiologist. When a radiologist assesses the original MR images, both the icobrain aria annotated images and the original, unaltered images are used for concurrent reading.

Intended patient population

icobrain aria is intended to be used in the population of patients with Alzheimer's disease receiving treatment with an amyloid betadirected antibody using the same ARIA radiographic severity scale as described in Aduhelm®/Leqembi® prescription information, and presenting for safety monitoring, including severity assessment and monitoring changes of ARIA.

Warnings and precautions

icobrain aria is an adjunct tool and is not intended to replace a radiologist's review of the images or his or her clinical iudgment. icobrain aria is not intended as an independent interpretation of MR images. Decisions should not be made solely based on analysis by icobrain aria.

The device is not intended to be used to segment macrohemorrhages with a diameter of 10 mm or more).

Similar to the predicate device (OsteoDetect, DEN180005), icobrain aria is intended to be used by radiologists interpreting radiological images, to help them with localizing anormalities. The devices are intended to be used concurrently with the reading of images and are not intended as a replacement for the review of a clinicial judgement. The indications for use of icobrain aria and the predicate device differ in the disease-specific findings the devices detect, the type of medical images the devices process, and the intended patient population. However, these differences should not raise new questions regarding safety and effectiveness of the device when used as labeled, under the same general and special controls as the predicate device.

6 Comparison of technological characteristics with the predicate device

At a high level, the subject and predicate devices are based on the following similar design and technological characteristics elements:

· software-only;
· the input consists of native DICOM images;
the end user only interacts with the electronic output of the device (including annotated DICOM images);
both devices employ machine learning (more specifically, supervised deep learning) methodology for detecting and diagnosing abnormalities in medical images.

The following technological differences exist between the subject and predicate devices:

different type of medical images the devices process;
annotated DICOM image output is a bounding box for OsteoDetect and color-coded 3D regions of interest for icobrain aria.

Both devices rely on highly similar forms of standalone performance testing and clinical performance testing to demonstrate effectiveness. Despite the different disease domains in which both devices are applied, icobrain aria was found to enable a comparable improvement in clinical performance to the intended users.

7 Performance data

7.1 Software verification and validation testing

Software verification and validation testing was conducted and documentation was provided as recommended by FDA's Guidance for Industry and FDA Staff, "Content of Premarket Submissions for Device Software Functions" (issued on June 14, 2023). Basic documentation level was applied to this submission.

10

Image /page/10/Picture/0 description: The image shows the logo for Icometrix. The logo consists of a stylized brain-like shape on the left, with the word "icometrix" in black, sans-serif font to the right of the shape. The brain shape is made up of three curved lines, colored in shades of blue, green, and red.

The device's software development, verification, and validation have been carried out following FDA-recognised standards and guidance documents, including:

IEC 62304:2006/Amd1:2015 Medical device software Software life cycle processes.
ISO 14971:2019 Medical devices Application of risk management to medical devices.
IEC 62366-1:2015/Amd1:2020 Medical devices Application of usability engineering to medical devices.
Computer-Assisted Detection Devices Applied to Radiology Device Data Premarket Notification [510(k)] Submissions, Document issued on September 28, 2022.
· Guidance for Industry and FDA Staff Clinical Performance Assessment: Considerations for Computer-Assisted Detection Devices Applied to Radiology Images and Radiology Device Data - Premarket Approval (PMA) and Premarket Notification [510(k)] Submissions, Document issued on September 28, 2022.
Off-The-Shelf Software Use in Medical Devices, Guidance for Industry and Food and Drug Administration Staff, Document issued on September 27, 2019.
Applying Human Factors and Usability Engineering to Medical Devices, Guidance for Industry and Food and Drug Administration Staff, Document issued on: February 3, 2016.
General Principles of Software Validation; Final Guidance for Industry and FDA Staff, Document issued on: January 11, 2002.
· DEN180005 Evaluation of automatic class III designation for OsteoDetect Decision summary with special controls.

The standalone software was tested against software design specific test plans to assure the device meets performance targets, as intended. The device risk management file was completed and risk controls were implemented to mitigate identified hazards. The testing results support that all the software specifications have met the individual modules and at system level. As such, icobrain aria passes all the testing and substantial equivalence with the predicate.

Validation activities also included a usability study (as an add-on questionnaire to the MRMC reader study described in section 7.6 Clinical Testing below), which demonstrated that intended users consider the device usable and useful, and the HCP manual provided with the device as comprehensive.

7.2 Biocompatibility testing

There are no direct or indirect patient components. Therefore, patient contact information is not needed for this device.

7.3 Electrical safety and electromagnetic compatibility (EMC)

icobrain aria is a software-only device, therefore electrical safety and EMC testing are not applicable.

7.4 Bench testing (Standalone Performance Testing)

icometrix conducted standalone performance assessments on cases that are representative for the indication for use in terms of patient demographics, diversity of ARIA findings and MRI scanner variability. Acceptance criteria were established based on statistical analyses of inter-expert variability for ARIA case-level and finding-level detection, which was complemented by a literature study on the radiological reading of ARA, and a review of CADe/CADx devices for other clinical applications and intended patient populations.

Data characteristics: The study data population consists of MRI datasets from subjects diagnosed with Alzheimer's disease. To guarantee independence of test data from training data, MR images used for testing were acquired from subjects that were not included in the training set. MRI scans originated from more than 100 sites in 20 countries, with approximately half the data originating from the US and the other half from outside the US. The protocols used to collect images match the required acquisition parameters of the finished software as intended for clinical use (cfr. HCP manual).

The main test set on which standalone performance results are reported coincides with the testing (see section below, where a direct comparison is made between performance of radiologists assisted by the software, and the standalone software itself). For testing the ARIA-H models, the dataset consisted of 199 pairs of MRI scans from 199 individual subjects. Mean (SD) age was 70.4 (7.2) years, ranging from 52 to 89; 105 (52.8%) were Asian, 1 (0.5%) was Black, 157 (78.9%) were White, and 18 (9.0%) were other or unreported race and ethnicity. The scans were acquired in 33 distinct MRI scanners manufactured by GE (26%), Philips (17%), at 1.5 Tesla (44%) and 3 Tesla (56%).

An extended test set was also available for additional standalone performance testing. This largerabhics characteristics as the main test set: mean (SD) age was 70.6 (7.57), ranging from 52 to 89; 53% female; 75% White, 12% Asian, 1% Black or African-American, and 12% not reported. The scans were acquired in MRI scanners manufactured by GE (24%), Philips (19%) and Siemens (57%), at 1.5 Tesla (42%) and 3 Tesla (58%). For testing the dataset consisted of 1015 pairs of MRI scans from 385 individual subjects. For testing the ARA-H models, the dataset consisted of 449 pairs of MRI scans from 369 individual subjects.

Methods: The standalone testing consisted of comparing software outputs against expert manual annotations. To establish the ground truth, expert neuroradiologists (with experience performing safety ARIA reading in clinical trials for Aβ-directed antibody

11

Image /page/11/Picture/0 description: The image shows the logo for Icometrix. The logo consists of a stylized, curved shape in shades of blue, green, and red on the left, followed by the word "icometrix" in a sans-serif font. The word is written in lowercase letters and is black.

therapies in AD) manually segmented both ARIA-H findings. Ground truth ARIA measurements were derived from the expert manual annotated masks.

The following main testing scenarios were considered:

1. Diagnostic performance testing for disting (on the case-level) between no ARIA-E and (mild, moderate or severe) ARIA-E, for distinguishing between no ARIA-H and (mild, moderate or severe) ARIA-H, and for distinguishing between severity levels for ARIA-E and ARIA-H, respectively, specificity and area under the receiver operating characteristic curve (AUC) are reported as mean and bootstrapped 95% confidence intervals. ARIA severity discrimination performance is evaluated in terms of weighted agreement to experts.
1. Detection performance, measured as (finding-level) detection rate, false positive rate and misclassification rate (the latter only for ARIA-H) for individual findings (ARIA-E sites of involvement, ARIA-H new microhemorrhages, ARIA-H new areas of superficial siderosis).

Results:

1. ARIA-E diagnostic performance in terms of case-level detection of no vs (mild, moderate or severe) ARIA-E was evaluated on a sample of 199 cases with 123 positive cases. The standalone software had a sensitivity of 0.94 (95% Cl [0.90, 0.98]], specificity 0.67 (95% Cl [0.56, 0.77]), and AUC 0.84 (95% Cl [0.77, 0.89]).
  ARIA-H diagnostic performance in terms of case-level detection of no vs (mild, moderate or severe) ARIA-H was evaluated on a sample of of 199 cases with 120 positive for ARIA-H microhemorrhages, 78 positive for ARIA-H superficial siderosis). The standalone software had a sensitivity of 0.87 (95% Cl [0.80, 0.93]], specificity 0.66 (95% Cl [0.55, 0.76]], and AUC 0.81 (95% Cl [0.75, 0.86]). Analogously, for ARA-H microhemorrhages, sensitivity was 0.89 (95% Cl [0.83, 0.92]), specificity was 0.62 (95% Cl [0.54, 0.67]), and AUC was 0.80 (95% Cl [0.74, 0.85]). For ARIA-H superficial siderosis, sensitivity was 0.67 (95% Cl [0.57, 0.76]), specificity was 0.95 (95% Cl [0.91, 0.98]), and AUC was 0.82 (95% Cl [0.76, 0.87]).

ARIA severity discrimination was sufficiently interchangeable with the severity ratings of ARIA experts, taking into account their mutual variability.

1. ARIA-E finding-level detection was evaluated on a sample of 199 cases with 123 positive cases. The standalone software had a finding-level true positive rate of 69.1% (95% Cl [62.4%, 75.5%]) with 0.7 (95% Cl [0.5, 0.8]) false positive findings per case.
  ARIA-H new microhemorrhages finding-level detection was evaluated on a sample of 199 cases with 82 positive cases. The standalone software had a finding-level true positive rate of 66.1% (95% Cl [57.7%, 74.1%)] with 0.9 (95% Cl [0.7, 1.1]) false positive findings per case.

ARIA-H new superficial siderosis finding-level detection was evaluated on a sample of 199 cases with 78 positive cases. The standalone software had a finding-level true positive rate of 62.5% (95% Cl [51.2%, 73.5%]) with 0.1 (95% Cl [0.1, 0.2]) false positive findings per case.

Misclassification between individual new microhemorrhages and new superficial siderosis findings was of the order of 0.2-0.3 findings per ARIA-H-positive case.

Additional performance testing: The additional performance testing on the extended test set and on both main and extended sets combined showed similar performance results as those evaluated on the main test set only. Generalizability was tested by subgroup analyses with respect to country of origin, expert ground truther, assessed ARA severity, MRI scanner (mis)match, etc.

Conclusions: The standalone performance testing results demonstrated that the software detects ARIA-E and ARIA-H at case level and individual finding level in line with the performance of human experts.

7.5 Animal testing

Not applicable. Animal studies are not necessary to establish the substantial equivalence of this device.

7.6 Clinical testing

icometrix conducted a fully-crossed multiple-case (MRMC) retrospective reader study in line with the special controls of 21 CFR 892.2090, and following recommendations from the FDA guidance document "Clinical Performance Assessment: Considerations for Computer-Assisted Devices Applied to Radiology Images and Radiology Device Data" (issued Sept 2022).

Methods: A multiple-reader multiple-case reader study was conducted to compare the diagnostic performance of radiologists assessing ARIA on MRI scans from patients treated with aducanumab in two situations: unassisted by the icobrain aria software.

The study encompassed 199 cases, and 16 U.S. Board of Radiologists with a range of experience. Radiologists reported the presence, severity and location of ARIA-H (microhemorrhages and superficial siderosis). Cases with ARA were over-represented in the sample for study design purposes. The co-primary study endpoints were the difference between assisted and unassisted detection of ARIA-H (considering in turn either microhemorrhages, superficial siderosis, or both) independently, assessed with the area under the empirical receiver operating characteristic curve (AUC), where the reader assessments for ARIA severity were compared against ground truth obtained via a consensus of 3 experts. Secondary endpoints evaluated the difference in diagnostic performance of detecting mild ARIA in the subgroup of cases with no and mild ARIA, and in discriminating moderate-or-severe ARA versus no-or-mild ARA, for ARIA-H (considering in turn either microhemorrhages,

12

Image /page/12/Picture/0 description: The image shows the logo for Icometrix. The logo consists of a stylized, abstract shape resembling a brain or a connected network, with a gradient of colors ranging from teal to green to red. To the right of the shape is the word "icometrix" in a clean, sans-serif font, with the "ico" portion in a lighter weight and the "metrix" portion in a bolder weight.

superficial siderosis, or both), independently. Accuracy in ARIA localization performance was also assessed. Finally, sensitivity and specificity for all endpoints, inter-reader variability, and reading time were also computed.

Results: Radiologists assisted by icobrain aria were significantly better in detecting ARA, with an average assisted AUC 0.873 (95% Cl [0.835, 0.911]) for ARIA-E detection, 0.825 (95% C [0.781, 0.869]) for ARIA-H (pooled microhemorrhages and superficial siderosis) detection (ARIA-H microhemorrhages: assisted AUC 0.808 (95% Cl [0.760, 0.855]); ARIA-H superficial siderosis: assisted AUC 0.784 (95% Cl [0.732, 0.836])). As such, icobrain aria-assisted reading performance was superior (vs unassisted) for all ARIA detection coprimary endpoints, with an AUC difference of 0.051 (95% Cl [0.020, 0.083]) for ARIA-E (p=0.001), 0.044 (95% Cl [0.017, 0.070]) for pooled ARIA-H (p=0.001), 0.029 (95% Cl [0.002, 0.055]) for ARIA-H microhemorrhages (p=0.032), and 0.063 (95% Cl [0.023, 0.102]) for ARIA-H superficial siderosis (p=0.003). Sensitivity increased significantly from 70.9% (unassisted) for ARIA-E detection, from 68.7% to 79.0% for pooled ARIA-H detection, from 69.3% to 79.6% for ARIA-H microhemorrhages detection, and from 49.7% to 59.9% for ARIA-H superficial siderosis detection, while specificity remained above 80% for the detection of all ARA types, except for a specificity of 76.7% for the detection of (mild) ARIA-H microhemorrhages.

Table: Primary endpoint results for ARIA detection performance. Results are reported for assisted and standalone software. Effect size is defined as the difference between assisted and unassisted AUC, and the p-value corresponds to the hypothesis test on the AUC difference.

| type | without:with
condition
ratio | effect size
[95% CI] | assisted | unassisted | p-value
(MRMC U
statistic
test) | software
standalone | |
|---------------------------------------------------------------------------------------------|------------------------------------|-------------------------|----------------------|------------|------------------------------------------|------------------------|-------|
| ARIA detection performance | | | | | | | |
| no ARIA-E vs. (mild, moderate or severe) ARIA-E | AUC | 76:123 cases | 0.051 [0.020, 0.083] | 0.873 | 0.822 | 0.001 | 0.838 |
| | sensitivity | | | 0.865 | 0.709 | | 0.943 |
| | specificity | | | 0.830 | 0.917 | | 0.671 |
| no ARIA-H vs. (mild, moderate or severe) ARIA-H | AUC | 79:120 cases | 0.044 [0.017, 0.070] | 0.825 | 0.781 | 0.001 | 0.811 |
| | sensitivity | | | 0.790 | 0.687 | | 0.867 |
| | specificity | | | 0.803 | 0.828 | | 0.658 |
| no ARIA-H microhemorrhages vs. (mild, moderate or severe) ARIA-H microhemorrhages | AUC | 117:82 cases | 0.029 [0.002, 0.055] | 0.808 | 0.779 | 0.032 | 0.802 |
| | sensitivity | | | 0.796 | 0.693 | | 0.890 |
| | specificity | | | 0.767 | 0.831 | | 0.624 |
| no ARIA-H superficial siderosis vs. (mild, moderate or severe) ARIA-H superficial siderosis | AUC | 121:78 cases | 0.063 [0.023, 0.102] | 0.784 | 0.721 | 0.003 | 0.816 |
| | sensitivity | | | 0.599 | 0.497 | | 0.667 |
| | specificity | | | 0.956 | 0.927 | | 0.950 |

ARIA detection performance

The highest gain in performance for the assisted reads was an increase from 47.2% to 70.2% in sensitivity for detecting mild ARIA-E (secondary endpoint).

Accuracy in localization performance for spatial distribution of both ARIA-H was also significantly better for the assisted reads compared to unassisted reads. The absolute differences between the individual ARIA severity measurements (i.e. ARIA-E longest axis, count of brain regions affected by ARA-E, count of new microhemorrhages and count of new superficial siderois sites), performed by the readers and the corresponding measurements derived from the expert ground truth masks were all significantly lower assisted versus unassisted.

The inter-reader variability was also significantly lower for assisted reads (Kendall's coefficient of concordance of 0.720 for unassisted and 0.809 for assisted ARIA-E severity assessments, and of 0.656 for unassisted ARIA-H severity assessments), and readers were on average faster (median unassisted reading 2:34min; median assisted reading 2:21min) when performing assisted reading with icobrain aria.

13

Image /page/13/Picture/0 description: The image shows the logo for icometrix. The logo consists of a stylized, curved shape in shades of blue, green, and red, resembling a brain. To the right of the shape is the word "icometrix" in a sans-serif font, with the letters in black.

Conclusions: Radiological reading performance for ARA detection and diagnosis is significantly improved when using icobrain aria. This demonstrates icobrain aria if used in clinical practice may enhance the sensitivity of safety monitoring of patients under amyloid beta-directed antibody therapies.

8 Conclusions

Non-clinical and clinical performance that icobrain aria is substantially equivalent to the chosen predicate. icobrain aria differs from the predicate device in the indications for use, because it detects different disease specific findings in different radiological images. However, both the proposed device (icobrain aria) and the predicate device (OsteoDetect) are computer-assisted detection and diagnostic devices that accept as input medical images in DICOM format and use machine learning techniques to identify and highlight image abnormalities. The overall design of the basic functionality that it provides to the end-user are the same. The differences in technological characteristics do not raise different questions of safety and effectiveness.

The risks associated with use of the device are comparable because they are both intended to be used in a similar same way as an concurrent reading aid for the radiologist, not as a replacement for their clinical judgement. Main risks are related to false positives and false negatives of the devices and these are mitigated by special controls defined in DEN180005. While the risks for both the subject and the predicate device are very similar, the icobrain aria device targets a different clinical condition compared to the predicate, potentially leading to different clinical harms. For this reason, a benefit/risk determination was conducted for icobrain aria:

| Summary of the
benefits | The clinical performance assessment multiple-reader multiple-case (MRMC) study demonstrated a
statistically significant improvement in reader performance in detecting ARIA-E and ARIA-H in the intended
patient population, as measured by the co-primary endpoints of the difference in Area Under the Curve for
assisted radiological reading with icobrain aria as compared to the same task without icobrain
aria according to clinical standard of care: |
|----------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| | ARIA-E detection: AUCassisted- AUCunassisted = 0.873 - 0.822 = 0.051 (two sided 95% confidence interval [0.020,
0.083]) |
| | ARIA-H detection: AUCassisted- AUCunassisted = 0.825 - 0.781= 0.044 (two sided 95% confidence interval
[0.017, 0.070]). |
| | The MRMC study also demonstrated that mean sensitivity increased significantly from 70.9% (unassisted)
to 86.5% (assisted) for ARIA-E detection, and from 68.7% to 79.0% for ARIA-H detection, while the device-
assisted specificity remained above 80% for the detection of both ARIA types. |
| | Large diagnostic gains included an increase in assisted versus unassisted performance for mild ARIA-E and
mild ARIA-H detection (AUCassisted- AUCunassisted = 0.071 for ARIA-E and AUCassisted- AUCunassisted = 0.042 for
ARIA-H), notably a much higher sensitivity, at no downside in performance for the easier tasks of detecting
moderate-or-severe ARIA-E and ARIA-H. Earlier detection of subtle ARIA findings will allow earlier
intervention and mitigate ARIA progression through appropriate decisions on dosing continuation or
suspension. |
| | The inter-reader variability in ARIA severity determination was also significantly lower for assisted versus
unassisted reads (Kendall's coefficient of concordance of 0.720 for unassisted and 0.809 for assisted ARIA-E
severity assessments, and of 0.656 for unassisted and 0.799 for assisted ARIA-H severity assessments), and
readers were on average faster (median unassisted reading 2:34min; median assisted reading 2:21min) when
performing assisted reading with icobrain aria. |
| Summary of the
risks | There are potential risks associated with use of the device, including:
• The device could provide false positive results (i.e., falsely detecting ARIA-E if there is no ARIA-E, false
detecting ARIA-H if there is no ARIA-H), which could contribute to the radiologist using this information
to make a false positive diagnosis (i.e., diagnosing ARIA-E when there is no ARIA-E, diagnosing ARIA-H
when there is no ARIA-H, overestimating mild ARIA-E or mild ARIA-H as moderate-or-severe ARIA-E or
ARIA-H, respectively). Such a false positive diagnosis can result in unnecessary additional MRI follow-up,
unnecessary suspension of amyloid beta-directed antibody treatment, delay in treatment (if false
positive findings correspond to symptoms caused by some other condition), HCP/patient concern,
inconvenience, discomfort from additional MRI, delay to potential benefit due to suspension of the
amyloid beta-directed antibody treatment,
• The device could provide false negative results (i.e., not detecting actual ARIA-E or ARIA-H findings),
which could contribute to the radiologist using this information to make a false negative diagnosis (e.g.,
not diagnosing ARIA-E when there is ARIA-E, not diagnosing ARIA-H when there is ARIA-H,
underestimating moderate-or-severe ARIA-E or ARIA-H as mild ARIA-E or ARIA-H, respectively). A false
negative diagnosis could lead to delayed diagnosis, delayed suspension of treatment when necessary,
symptoms exacerbating, no follow-up MRI,
• The device could be misused to analyze brain images from other patient populations or incompatible
image acquisition parameters, leading to inappropriate information regarding the presence/location of
imaging abnormalities provided to the end user.
• The device could fail and lead to absence of results, delay of results, or incorrect results, which can lead
to delayed or inaccurate patient diagnosis. |

14

Image /page/14/Picture/0 description: The image shows the logo for icometrix. The logo consists of a stylized brain-like shape on the left, with the word "icometrix" in black sans-serif font to the right of the shape. The brain shape is colored with a gradient, transitioning from blue at the top to green in the middle and red at the bottom.

| Benefit-risk
considerations
(Do the probable
benefits outweigh
the probable
risks?) | The assistive icobrain aria software aims to support radiologists in detecting ARIA and grading its severity
for reporting to referral clinicians. Such tool contributes to improved patient wellbeing and outcome
through improved clinical decisions and workflow. icobrain aria is not a replacement of the standard clinical
practice but rather a supplement to it. In case of clinical decisions based on incorrect data inputs or
incorrect data interpretation by the practitioners, there will be no direct significant impact on the patient's
health, as potential harm can be handled by regular clinical and MRI monitoring. |
|----------------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| | The benefited population would include Alzheimer's disease patients receiving an amyloid beta-directed
antibody treatment and monitored for symptomatic or asymptomatic ARIA. Based on the performance data
and the application of general controls and special controls in accordance to the device classification and in
alignment with the predicate device, it is unlikely that the aforementioned risks would result in worse ARIA
monitoring compared with the current clinical standard of practice. Clinical performance assessment
reveals that there is little to no risk of worse health outcome due to automation bias (i.e., users accepting the
suggested findings by the software without critical review). This software device potentially brings general
radiologists, who have little to no experience with ARIA monitoring, on par with experts that participated in
ARIA safety monitoring in clinical trials of amyloid beta-directed antibody therapies. Possible misuse of the
device does not present additional risks compared with the misuse of other types of radiological image
processing devices. Use of incompatible image acquisition parameters is mitigated by not providing the
icobrain aria outputs to the end user if non-compliant protocols are used. |
| | The device has been developed and tested on brain scans acquired from Alzheimer's disease patients who
participated in the clinical trials of aducanumab, and has not yet been extensively tested on other patient
populations. Nevertheless, ARIA is a radiographic finding that, when it occurs, has identical appearance for
other amyloid beta-directed antibody treatments, which should not limit applicability of the device to only
aducanumab-treated patients. However, the sample population on which the device was tested included
only patients who qualified to the inclusion criteria of clinical trials and might not represent all possible
concomitant pathologies that may occur in parallel to Alzheimer's disease. Since more data is currently
unavailable for product development outside clinical trials, post-market vigilance criteria will be applied to
monitor potentially higher rates of false positives or false negatives in real world intended population. |
| Conclusion | icometrix has applied a risk management process in accordance with FDA recognized standards to identify,
evaluate, and mitigate all known hazards related to icobrain aria. These hazards may occur when accuracy
of diagnosis is potentially affected, causing either false positives or false negatives. All identified risks are
effectively mitigated and it can be concluded that the residual risk is outweighed by the benefits. |

Because of similar intended use and technological characteristics, icobrain aria is substantially equivalent to a device (OsteoDetect, DEN180005) that has been granted its De Novo request in the United and clinical and clinical performance tests demonstrate that icobrain aria is as safe and effective as the predicate device.