(372 days)
Deep Capsule® is an artificial intelligence (AI) assisted reading tool designed to aid small bowel capsule endoscopy reviewers in decreasing the time to review capsule endoscopy images for adult patients in whom the capsule endoscopy images were obtained for suspected small bowel bleeding. The clinician is responsible for conducting their own assessment of the findings of the AI-assisted reading through review of the entire video, as clinically appropriate. This tool is not intended to replace clinical decision-making.
Deep Capsule® is an artificial intelligence (AI) assisted reading tool designed to aid small bowel capsule endoscopy reviewers (SBCapER) in decreasing the time to review capsule endoscopy images for adult patients in whom the capsule endoscopy images were obtained for suspected small bowel bleeding.
Deep Capsule® is capable of detecting small bowel lesions, without differentiating them. The detection of lesions by Deep Capsule® is insufficient to achieve a direct diagnosis, which is dependent on the clinical integration of the different findings. Deep Capsule® should be integrated into this multifactorial and complex context, both in the diagnostic workup or in the patient follow-up.
In summary, although the Deep Capsule® detects small bowel lesions, it acts only as a support to the clinical decision. The ultimate diagnosis will always be given by the small bowel capsule endoscopy reviewers ("human in the loop").
It is important to note that Deep Capsule® also provides in the findings count as a non-AI software output, and user inputs that are inserted and edited by SBCapERs, including CapE device model, patient exam priority category, procedure date, responsible physician, clinical indication and medical notes.
Deep Capsule® software includes a main algorithm as illustrated in Figure 1:
• Small bowel lesion detection, which includes the small bowel lesion image analysis algorithm designed to automatically identify and localize potential small bowel lesions in capsule endoscopy images. Suspected findings are selected automatically from the video frames as illustrated in Figure 1 below, with the active video displayed on the left side of the user interface. On the right side, AI-selected frames are presented as a structured gallery of thumbnails, allowing the reviewer to efficiently navigate through suspected findings. The interface also provides exam details, frame indexing, clinical indication, findings count, and medical notes to support structured review.
Here's a breakdown of the acceptance criteria and the study proving the device meets them, based on the provided FDA 510(k) clearance letter for Deep Capsule®:
1. Table of Acceptance Criteria and Reported Device Performance
The FDA clearance letter does not explicitly define "acceptance criteria" for the device's performance in a table format. However, it presents the results of its performance studies against benchmarks like "Expert Reading" (standalone algorithm testing) and "Expert Board (Ground Truth)" (clinical validation).
Based on the provided data, we can infer the performance metrics deemed acceptable:
| Metric (Inferred Acceptance Criteria) | Reported Device Performance (Deep Capsule®) | Study Phase |
|---|---|---|
| Standalone Algorithm Testing (Lesion Detection Function) | ||
| Patient-level Sensitivity | 95.8% (95% CI: 92.7%-97.9%) | Internal Test Set |
| Patient-level Specificity | 75% (95% CI: 34.9%-96.8%) | Internal Test Set |
| Image-level Sensitivity | 92.1% (95% CI: 91.9%-92.3%) | Internal Test Set |
| Image-level Specificity | 88.0% (95% CI: 87.4%-88.6%) | Internal Test Set |
| Clinical Validation Study (AI-aided CapE vs. SoC vs. Expert Board) | ||
| Diagnostic Yield (Non-inferiority to SoC) | 0.961 (0.934 - 0.977) for AI-aided CapE, 0.761 (0.712 - 0.803) for SoC; non-inferiority established (p < 0.001) | Clinical Validation |
| Per-Patient Sensitivity (AI-aided CapE) | 0.972 (0.947 - 0.986) | Clinical Validation |
| Per-Patient Specificity (AI-aided CapE) | 0.125 (0.055 - 0.261) | Clinical Validation |
| Per-Patient PPV (AI-aided CapE) | 0.890 (0.850 - 0.920) | Clinical Validation |
| Per-Patient NPV (AI-aided CapE) | 0.385 (0.177 - 0.645) | Clinical Validation |
| Image-level Sensitivity (AI-aided CapE) | 94% (89.6 - 96.9) | Clinical Validation |
| Image-level Specificity (AI-aided CapE) | 84.9% (82.6 - 87.2) | Clinical Validation |
| Image-level PPV (AI-aided CapE) | 27% (21.5 - 32.9) | Clinical Validation |
| Image-level NPV (AI-aided CapE) | 99.6% (99.3 - 99.8) | Clinical Validation |
| Reduction in Mean Reading Time with AI-aided reading | "significantly reduced" (quantification not provided) | Clinical Validation |
2. Sample Size for the Test Set and Data Provenance
-
Standalone Algorithm Testing Test Set:
- Patients: 272 patients (from a total dataset of 1,133 patients).
- Images: 101,802 images (from a total dataset of 321,357 images).
- Data Provenance: Retrospectively collected from two clinical institutions in Portugal (Centro Hospitalar Universitário São João, Porto, Portugal; and ManopH – Laboratório de Endoscopia e Motilidade Digestiva, Lda., Vila Nova de Gaia, Portugal).
-
Clinical Validation Study Test Set:
- Patients: 330 patients.
- Data Provenance: Retrospectively collected between January 2021 and April 2024 from seven independent clinical centers across four countries: Portugal, Spain, Brazil, and the United States. No overlap existed between training and clinical validation datasets, and clinical validation sites were independent from the institutions used for model development.
3. Number of Experts Used to Establish Ground Truth for the Test Set and Qualifications
- Standalone Algorithm Testing (Internal Test Set): The document states that images were "expert-labeled images." It does not specify the exact number of experts, their qualifications, or the adjudication method for this initial labeling of the training/internal test dataset.
- Clinical Validation Study (Ground Truth):
- Adjudicators: 5 independent expert adjudicators.
- Qualifications: "board-certified gastroenterologists with extensive capsule endoscopy experience."
- Clinical Readers: 15 board-certified gastroenterologists.
- Qualifications mentioned for Clinical Readers: "certified specialists in gastroenterology and/or digestive endoscopy with a minimum of 5 years of post-fellowship clinical experience." Readers were located across the United States, Portugal, Spain, and Brazil. There was no overlap between clinical readers and adjudicators.
- Image-level ground truth: "established by expert reviewers blinded to AI outputs." (Implies the 5 expert adjudicators).
4. Adjudication Method for the Test Set
- Standalone Algorithm Testing (for initial expert-labeled images - training/internal test): Not explicitly stated, but implies expert consensus or single expert labeling for the "expert-labeled images."
- Clinical Validation Study (for Expert Board Ground Truth): "Ground truth was established through independent manual review of full capsule endoscopy videos by two experienced gastroenterologists, with adjudication by an expert board in cases of disagreement." This is a 2+1 adjudication model.
5. Multi Reader Multi Case (MRMC) Comparative Effectiveness Study
Yes, an MRMC comparative effectiveness study was done.
- Effect Size of Human Readers Improving with AI vs. without AI assistance:
- Diagnostic Yield: AI-aided CapE showed a Diagnostic Yield (DY) of 0.961 compared to Standard-of-Care (SoC) at 0.761. The difference in DY between AI-aided and SoC was 0.200 (0.149 - 0.251), with non-inferiority established (p < 0.001). This indicates a substantial improvement in diagnostic yield when readers are assisted by AI.
- Sensitivity (Per-Patient): AI-aided CapE had a sensitivity of 0.972 compared to SoC at 0.762. The improvement in sensitivity for AI-aided reading over unaided (SoC) was 21.0% (13.9 - 27.6%).
- Specificity (Per-Patient): AI-aided CapE had a specificity of 0.125 compared to SoC at 0.250. The difference in specificity was -12.5% (-34.7 - 11.9%). It decreased with AI assistance (due to the AI prioritizing sensitivity and detecting more potential lesions, leading to more "false positives" that clinicians must then review for true positivity).
- Mean Reading Time: "mean reading time was significantly reduced with AI+Physician reading, as compared to standard reading." (Specific quantification of reduction not provided in the excerpt).
6. Standalone (Algorithm Only Without Human-in-the-Loop Performance) Study
Yes, a standalone algorithm performance study was done. This is referred to as "Standalone Algorithm Testing" or "PERFORMANCE TESTING – BENCH – STANDALONE PERFORMANCE."
- Patient-Level Performance: Sensitivity 95.8%, Specificity 75%.
- Image-Level Performance: Sensitivity 92.1%, Specificity 88.0%.
7. Type of Ground Truth Used
- Standalone Algorithm Testing: "Expert-labeled images."
- Clinical Validation Study: Expert board reference standard involving "independent manual review of full capsule endoscopy videos by two experienced gastroenterologists, with adjudication by an expert board in cases of disagreement."
8. The Sample Size for the Training Set
- Patients: 861 patients (from a total of 1,133 patients).
- Images: 219,555 images (from a total of 321,357 images).
9. How the Ground Truth for the Training Set Was Established
The document states: "Images included in the dataset [which includes the training set] were labeled by expert readers following a structured annotation process." It does not provide further details on the number of experts, their qualifications, or the exact adjudication method during the training data labeling phase.
FDA 510(k) Clearance Letter - Deep Capsule®
Page 1
U.S. Food & Drug Administration
10903 New Hampshire Avenue
Silver Spring, MD 20993
www.fda.gov
Doc ID # 04017.08.03
March 12, 2026
DigestAID - Artificial Intelligence Development, S.A.
Miguel Mascarenhas
CEO
Rua Particular Carlos Fontes, 117, São Cosme
Gondomar, Porto 4420-249
Portugal
Re: K250655
Trade/Device Name: Deep Capsule® (Deep Capsule US)
Regulation Number: 21 CFR 876.1540
Regulation Name: Gastrointestinal Capsule Endoscopy Analysis Software Device
Regulatory Class: Class II
Product Code: QZF
Dated: March 10, 2026
Received: March 10, 2026
Dear Miguel Mascarenhas:
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"
Page 2
(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 Management System Regulation (QMSR) (21 CFR Part 820), which includes, but is not limited to, ISO 13485 clause 7.3 (Design controls), ISO 13484 clause 8.3 (Nonconforming product), and ISO 13485 clause 8.5 (Corrective and preventative action). Please note that regardless of whether a change requires premarket review, the QMSR requires device manufacturers to review and approve changes to device design and production (ISO 13485 clause 7.3 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 (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-reporting-combination-products); good manufacturing practice requirements as set forth in the Quality Management System Regulation (QMSR) (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 Rule"). 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-devices/device-advice-comprehensive-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-devices/medical-device-safety/medical-device-reporting-mdr-how-report-medical-device-problems.
For comprehensive regulatory information about medical devices and radiation-emitting products, including information about labeling regulations, please see Device Advice (https://www.fda.gov/medical-devices/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-devices/device-advice-comprehensive-regulatory-assistance/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).
Page 3
Sincerely,
SHANIL P. HAUGEN -S
Shanil P. Haugen, Ph.D.
Assistant Director
DHT3A: Division of Renal, Gastrointestinal,
Obesity and Transplant Devices
OHT3: Office of Gastrorenal, ObGyn,
General Hospital and Urology Devices
Office of Product Evaluation and Quality
Center for Devices and Radiological Health
Enclosure
Page 4
DEPARTMENT OF HEALTH AND HUMAN SERVICES
Food and Drug Administration
Form Approved: OMB No. 0910-0120
Expiration Date: 07/31/2026
Indications for Use
See PRA Statement below.
510(k) Number (if known)
K250655
Device Name
Deep Capsule®
Indications for Use (Describe)
Deep Capsule® is an artificial intelligence (AI) assisted reading tool designed to aid small bowel capsule endoscopy reviewers in decreasing the time to review capsule endoscopy images for adult patients in whom the capsule endoscopy images were obtained for suspected small bowel bleeding. The clinician is responsible for conducting their own assessment of the findings of the AI-assisted reading through review of the entire video, as clinically appropriate. This tool is not intended to replace clinical decision-making.
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.
This section applies only to requirements of the Paperwork Reduction Act of 1995.
DO NOT SEND YOUR COMPLETED FORM TO THE PRA STAFF EMAIL ADDRESS BELOW.
The burden time for this collection of information is estimated to average 79 hours per response, including the time to review instructions, search existing data sources, gather and maintain the data needed and complete and review the collection of information. Send comments regarding this burden estimate or any other aspect of this information collection, including suggestions for reducing this burden, to:
Department of Health and Human Services
Food and Drug Administration
Office of Chief Information Officer
Paperwork Reduction Act (PRA) Staff
PRAStaff@fda.hhs.gov
"An agency may not conduct or sponsor, and a person is not required to respond to, a collection of information unless it displays a currently valid OMB number."
FORM FDA 3881 (8/23) Page 1 of 1
Page 5
510(k) Summary
In accordance with 21 CFR 807.92 the following summary of information is provided:
807.92(a)(1) – Submitter Information
| Field | Information |
|---|---|
| Date | 08-August-2025 |
| Submitter | DigestAID - Artificial Intelligence Development SA |
| Primary Contact Person | Dr. Miguel MascarenhasCEODigestAIDEmail: miguel.mascarenhas@digestaid.healthPh: +351 915 065 586 |
| Secondary Contact Person | Dr. Hélder CardosoCMODigestAIDEmail: helder.cardoso@digestaid.healthPh: +351 916 022 457 |
807.92(a)(2) – Device Information
| Field | Information |
|---|---|
| Device Trade Name | Deep Capsule® |
| Common/Usual Name | Gastrointestinal capsule endoscopy analysis software device |
| Regulation Name | Gastrointestinal capsule endoscopy analysis software device |
| Regulation Number | 876.1540 |
| Regulation Class | Class II |
| Product Code | QZF |
| Review Panel | Gastroenterology/Urology |
807.92(a)(3) – Predicate Device
| Field | Information |
|---|---|
| De Novo Number | DEN230027 |
| Manufacturer | Ankon Technologies Co., Ltd |
| CFR | 21 CFR § 876.1540 |
| Classification Product Code | QZF |
| Predicate Device(s) | NaviCam ProScan |
Page 6
Indications for Use
The Deep Capsule® is indicated as follows:
Deep Capsule® is an artificial intelligence (AI) assisted reading tool designed to aid small bowel capsule endoscopy reviewers in decreasing the time to review capsule endoscopy images for adult patients in whom the capsule endoscopy images were obtained for suspected small bowel bleeding. The clinician is responsible for conducting their own assessment of the findings of the AI-assisted reading through review of the entire video, as clinically appropriate. This tool is not intended to replace clinical decision-making.
Limitations
Deep Capsule® is not intended to be used as a stand-alone diagnostic device or replace clinical decision-making. The device is intended to assist qualified healthcare professionals in reviewing small bowel capsule endoscopy videos.
Deep Capsule® should only be use with compatible small bowel capsule endoscopy systems as specified in the labeling.
Deep Capsule® is not intended for autonomous diagnostic use. The clinician is responsible for making the final clinical diagnosis.
A negative or normal result determined by Deep Capsule® alone does not exclude the presence of small bowel disease (false negative). Similarly, a positive or abnormal result does not automatically confirm the presence of small bowel disease (false positive). The clinician should always carefully review the complete capsule endoscopy video in accordance with standard clinical practice. If clinical symptoms persist, further evaluation should be performed.
Deep Capsule® is not intended to characterize lesions in a manner that would replace biopsy sampling or other characterization tools.
PLEASE REFER TO THE LABELING FOR A COMPLETE LIST OF WARNINGS, PRECAUTIONS, AND CONTRAINDICATIONS.
Device Description
Deep Capsule® is an artificial intelligence (AI) assisted reading tool designed to aid small bowel capsule endoscopy reviewers (SBCapER) in decreasing the time to review capsule endoscopy images for adult patients in whom the capsule endoscopy images were obtained for suspected small bowel bleeding.
Deep Capsule® is capable of detecting small bowel lesions, without differentiating them. The detection of lesions by Deep Capsule® is insufficient to achieve a direct diagnosis, which is dependent on the clinical integration of the different findings. Deep Capsule® should be integrated into this multifactorial and complex context, both in the diagnostic workup or in the patient follow-up.
In summary, although the Deep Capsule® detects small bowel lesions, it acts only as a support to the clinical decision. The ultimate diagnosis will always be given by the small bowel capsule endoscopy reviewers ("human in the loop").
Page 7
It is important to note that Deep Capsule® also provides in the findings count as a non-AI software output, and user inputs that are inserted and edited by SBCapERs, including CapE device model, patient exam priority category, procedure date, responsible physician, clinical indication and medical notes.
Deep Capsule® software includes a main algorithm as illustrated in Figure 1:
• Small bowel lesion detection, which includes the small bowel lesion image analysis algorithm designed to automatically identify and localize potential small bowel lesions in capsule endoscopy images. Suspected findings are selected automatically from the video frames as illustrated in Figure 1 below, with the active video displayed on the left side of the user interface. On the right side, AI-selected frames are presented as a structured gallery of thumbnails, allowing the reviewer to efficiently navigate through suspected findings. The interface also provides exam details, frame indexing, clinical indication, findings count, and medical notes to support structured review.
The Deep Capsule® detection algorithm is based on convolutional neural networks using different deep learning models.
Page 8
Table 1: Comparison of Indications for Use and Technological Characteristics
| Characteristic | Subject Device | Predicate Device | Explanation of Difference |
|---|---|---|---|
| Device name | Deep Capsule® | NaviCam ProScan | N/A |
| FDA Clearance | K250655 | DEN230027 | N/A |
| Classification Name | Gastrointestinal capsule endoscopy analysis software device | Gastrointestinal capsule endoscopy analysis software device | Same |
| FDA Reg # | 21 CFR 876.1540 | 21 CFR 876.1540 | Same |
| FDA Product Code | QZF | QZF | Same |
| Classification | Class II | Class II | Same |
| Indications for Use | Deep Capsule® is an artificial intelligent (AI) assisted reading tool designed to aid small bowel capsule endoscopy reviewers in decreasing the time to review capsule endoscopy images for adult patients in whom the capsule endoscopy images were obtained for suspected small bowel bleeding. The clinician is responsible for conducting their own assessment of the findings of the AI-assisted reading through a review of the entire video, as clinically appropriate. This tool is not intended to replace clinical decision-making. | NaviCam ProScan is an artificial intelligent (AI) assisted reading tool designed to aid small bowel capsule endoscopy reviewers in decreasing the time to review capsule endoscopy images for adult patients in whom the capsule endoscopy images were obtained for suspected small bowel bleeding. The clinician is responsible for conducting their own assessment of the findings of the AI-assisted reading through review of the entire video, as clinically appropriate. ProScan also assists small bowel capsule endoscopy reviewers in identifying the digestive tract location (oral cavity and beyond, esophagus, stomach, | Similar - Both devices are computer-aided detection (CADe) tools designed to assist endoscopy reviewers in decreasing the time to review capsule endoscopy images for adult patients in whom the capsule endoscopy images were obtained for suspected small bowel bleeding. None of the technologies are intended to replace clinical decision-making. The only difference is that NaviCam ProScan also assists small bowel |
Page 9
| Characteristic | Subject Device | Predicate Device | Explanation of Difference |
|---|---|---|---|
| small bowel) of the image in adults. This tool is not intended to replace clinical decision-making. | capsule endoscopy reviewers in identifying the digestive tract location (oral cavity and beyond, esophagus, stomach, small bowel) of the image in adults, and Deep Capsule® does not. | ||
| Anatomical site | Small bowel | Small bowel | Same |
| Function | CADe | CADe | Same |
| Analysis | Capsule endoscopy images review | Capsule endoscopy images review | Same |
| Target Population | Adult patients with suspicion of small bowel bleeding. | Adult patients with suspicion of small bowel bleeding. | Same |
| Targeted Disease | Small bowel bleeding | Small bowel bleeding | Same |
| Type of Procedure | Capsule endoscopy | Capsule endoscopy | Same |
| Technological Characteristics | Deep Capsule® is a SaMD designed to detect potential small bowel lesions. | NaviCam ProScan is a SaMD designed to detect potential small bowel lesions and digestive tract structures location identification. | Similar - Both devices are SaMD to detect potential small bowel lesions. The only difference is that NaviCam ProScan has an extra feature for digestive tract location identification, while Deep Capsule® does not. |
Page 10
| Characteristic | Subject Device | Predicate Device | Explanation of Difference |
|---|---|---|---|
| Software Algorithm | Deep Capsule® uses an artificial intelligence-based algorithm to perform potential small bowel lesions detection. | NaviCam ProScan uses an artificial intelligence-based algorithm to detect potential small bowel lesions and identify digestive tract locations. | Similar - Both devices use an artificial intelligence-based algorithm to perform the detection of lesions in the small bowel. The only difference is that NaviCam ProScan also identifies digestive tract location, while Deep Capsule® does not. |
| Explainable AI Strategy | Is based on potential small bowel lesion frames detection/selection. | Is based on potential small bowel lesions frames detection/ selection. It also gives digestive tract structure locations. | Similar - Both devices are based on an artificial intelligence algorithm that selects potential lesion frames in the small bowel. The only difference is that NaviCam ProScan AI also identifies digestive tract location, while Deep Capsule® does not. |
| Device AI Output | Deep Capsule® generates a report with the potential small bowel lesion frames selected by the small bowel capsule endoscopy reviewer. | NaviCam ProScan generates a report with the potential small bowel lesion frames selected by the small bowel capsule endoscopy reviewer. It also gives digestive tract structure locations. | Similar - generates a report with the potential small bowel lesion frames selected by the small bowel capsule endoscopy reviewer. The only difference is that NaviCam ProScan also identifies the location of digestive tract structures, while Deep Capsule® does not. |
Page 11
| Characteristic | Subject Device | Predicate Device | Explanation of Difference |
|---|---|---|---|
| Device non-AI Output | Finding countPatient Exam Priority | Not applicable | The NaviCam ProScan is an AI tool integrated into ESView, the image review software used in the NaviCam SB System. The non-AI outputs are not part of the NaviCam ProScan tool. |
Page 12
Summary of Non-Clinical Tests/Bench Studies
Software/Cybersecurity
Deep Capsule® was identified as having a basic level of concern as defined in the FDA guidance document "Guidance for the Content of Premarket Submissions for Software Contained in Medical Devices." The software documentation included:
- Software Description
- Risk Management File
- Software Requirement Specification
- System and Software Architecture Design
- Software Design Specification
- Software Development, Configuration Management, and Maintenance Practices
- Software Testing as Part of Verification and Validation
- Software Version History
- Unresolved Software Anomalies
Risk analysis was provided for the software with a description of the hazards, their causes and severity as well as acceptable methods for control of the identified risks. Deep Capsule® provided a description, with test protocols including pass/fail criteria and report of results, of acceptable verification and validation activities at the unit, integration and system level. All testing met design specifications and passed successfully. This testing is not part of the AI performance evaluation.
Regarding the cybersecurity, documentation included recommended information from the FDA guidance document "FDA Cybersecurity in Medical Devices: Quality System Considerations and Content of Premarket Submissions" This includes threat identification, vulnerability assessment, likelihood and impact assessment, cybersecurity mitigation information, security policies and controls, continuous monitoring and review activities, regular auditing and cybersecurity testing.
PERFORMANCE TESTING – BENCH – STANDALONE PERFORMANCE
The optimization, training, and validation of the Deep Capsule® was performed in several phases, as summarized below.
Algorithm Training
In the training phase, Deep Capsule® was trained to recognize abnormal small bowel capsule endoscopy images for lesion detection. A dataset of 1,133 patients who underwent small bowel capsule endoscopy was collected from multiple Capsule Endoscopy System (PillCam™ SB3, PillCam™ Crohn, PillCam™ SB1, PillCam™ Colon2, OMOM® HD System, Olympus ENDOCAPSULE 10 (EC-10)) obtained from 2 clinical institutions in Portugal. The dataset included 321,357 expert-labeled images drawn from full-length capsule endoscopy examinations.
Of those 1,133 patients, the dataset was split so that:
● 861 patients (76.0%) were used for training the lesion detection function; and
Page 13
● 272 patients (24.0%) were used for internal testing of the lesion detection function (see Standalone Algorithm Testing with full demographics tables 2 and 3).
Table 2: Lesion Detection Function.
| Training (No. of subjects = 861) | Model (No. of subjects = 272) | |
|---|---|---|
| Age: | ||
| Mean Age, years (SDV) | 53.36 (17.90) | 58.11 (18.88) |
| Age range | 18–91 | 18–91 |
| Sex: | ||
| Male | 416 | 119 |
| Female | 445 | 153 |
| Race/Ethnicity: | ||
| White or Caucasian | 861 | 272 |
| Black or African American | 0 | 0 |
| Hispanic or Latino | 0 | 0 |
| Asian | 0 | 0 |
| Native Hawaiian or Other Pacific Islander | 0 | 0 |
| Other / Not Reported | 0 | 0 |
Table 3: Demographics of standalone performance dataset.
| Model Dataset | Model Training Set | Model Test Set | |
|---|---|---|---|
| Images | |||
| Images, N (%) | 321357 (100%) | 219555 (68.3%) | 101802 (31.7%) |
| Images per device (count) | |||
| PillCam™ SB3 | 229467 (71.41%) | 154200 (70.23%) | 75267 (73.93%) |
| PillCam™ Crohn | 71164 (22.14%) | 53027 (24.15%) | 18137 (17.82%) |
| OMOM® HD | 20619 (6.42%) | 12221 (5.57%) | 8398 (8.25%) |
| Olympus EC-10® | 45 (0.01%) | 45 (0.02%) | -- (0%) |
| PillCam™ Colon2 | 35 (0.01%) | 35 (0.02%) | -- (0%) |
| PillCam™ SB1 | 20 (0.01%) | 20 (0.01%) | -- (0%) |
| Subjects | |||
| Exams, N (%) | 1133 (100%) | 861 (75.9%) | 272 (24.1%) |
| Male, N (%) | 535 (47%) | 416 (48%) | 119 (44%) |
| Female, N (%) | 598 (53%) | 445 (52%) | 153 (56%) |
Page 14
| Model Dataset | Model Training Set | Model Test Set | |
|---|---|---|---|
| Mean Age, years (SDV) | 54.87 (18.35) | 53.36 (17.90) | 58.11 (18.88) |
| Subjects per device (count) | |||
| PillCam™ SB3 | 888 (78.24%) | 676 (78.3%) | 212 (77.9%) |
| PillCam™ Crohn | 82 (7.22%) | 82 (9.5%) | 20 (7.35%) |
| OMOM® HD | 137 (12.07%) | 97 (11.2%) | 40 (14.7%) |
| Olympus EC-10® | 1 (0.09%) | 1 (0.1%) | -- (0%) |
| PillCam™ Colon2 | 3 (0.26%) | 3 (0.4%) | -- (0%) |
| PillCam™ SB1 | 2 (0.18%) | 2 (0.23%) | -- (0%) |
| Subjects per clinical indication (count) | |||
| NA | 411 (36.27%) | 305 (35.42%) | 106 (38.97%) |
| Anemia | 281 (24.80%) | 225 (26.13%) | 56 (20.59%) |
| Crohn"s Disease Staging | 164 (14.47%) | 126 (14.63%) | 38 (13.97%) |
| Crohn"s Disease Diagnose | 133 (11.74%) | 100 (11.61%) | 33 (12.13%) |
| OGIB | 74 (6.53%) | 50 (5.81%) | 24 (8.82%) |
| FAP | 26 (2.29%) | 22 (2.56%) | 4 (1.47%) |
| Other Polyposis | 12 (1.06%) | 9 (1.04%) | 3 (1.10%) |
| NET | 8 (0.71%) | 7 (0.81%) | 1 (0.37%) |
| Peutz-Jeghers Syndrome | 7 (0.62%) | 4 (0.46%) | 3 (1.10%) |
| Common Variable Immunodefency | 6 (0.53%) | 4 (0.46%) | 2 (0.74%) |
| Celiac Disease | 3 (0.26%) | 3 (0.35%) | -- (0%) |
| Imagiological Anormality | 3 (0.26%) | 3 (0.35%) | -- (0%) |
| Chronic Diarrhea | 2 (0.18%) | 2 (0.23%) | -- (0%) |
| HNPCC | 2 (0.18%) | -- (0%) | 2 (0.74%) |
| CRC Screening | 1 (0.08%) | 1 (0.12%) | -- (0%) |
| Subjects per Race/Ethnicity | |||
| White or Caucasian | 1133 (100%) | 861 (100%) | 272 (100%) |
| Black or African American | -- (0%) | -- (0%) | -- (0%) |
| Hispanic or Latino | -- (0%) | -- (0%) | -- (0%) |
| Asian | -- (0%) | -- (0%) | -- (0%) |
| Native Hawaiian or other Pacific Islander | -- (0%) | -- (0%) | -- (0%) |
Note: Additionally, center-based stratification is not presented separately, as the training dataset was derived from two Portuguese centers: Centro Hospitalar Universitário São João (CHUSJ), Porto, Portugal (n = 1008 patients), and ManopH – Laboratório de Endoscopia e Motilidade Digestiva, Lda., Vila Nova de Gaia, Portugal (n = 127 patients).
Image Labeling Process
Images included in the dataset were labeled by expert readers following a structured annotation process.
Page 15
Capsule Compatibility and Acquisition System Transparency
The table below summarizes capsule system inclusion across development and validation phases:
Table 4: Standalone Performance, and Clinical Validation by Capsule Model
| Capsule Model | Standalone Performance | Clinical Validation |
|---|---|---|
| PillCam™ SB3 | Yes | Yes (n=307) |
| PillCam™ SB2 | No standalone | Limited (n=2) |
| ENDOCAPSULE 10 | No standalone | Yes (n=21) |
| OMOM® HD | Yes (internal) | No |
Performance metrics for certain capsule systems (e.g., PillCam™ SB2) are based on limited sample sizes and should be interpreted accordingly. OMOM® HD was included during internal development and standalone robustness testing but is not part of the U.S. compatibility claims.
This structured reporting ensures transparency regarding tested hardware and associated subgroup analyses, consistent with 21 CFR 876.1540 Special Control 5(ii).
Standalone Algorithm Testing
- Lesion detection function
Following the training phase, the performance of the Deep Capsule® to correctly recognize small bowel images with lesions was tested. For this purpose, 272 patients from the dataset described in Table 3 were selected for testing, with patient-level results provided in Tables 5 and 6 below. From that same dataset, normal and abnormal small bowel capsule endoscopy images were used to test the algorithm, with image-level results provided in Tables 7 and 8.
a. Patient-Level Analysis
Table 5: Patient-level Analysis Results.
| Deep Capsule Prediction | Expert Reading | Sensitivity (95% CI) | Specificity (95% CI) | AUC | |
|---|---|---|---|---|---|
| Lesion | Normal | ||||
| Lesion | 253 | 2 | 0.958 (0.927, 0.979) | 0.750 (0.349, 0.968) | 0.966 (0.939 - 0.993) |
| Normal | 11 | 6 |
Patient-level sensitivity and specificity were determined to be 95.8% (95% CI: 92.7%-97.9%) and 75% (95%CI: 34.9%-96.8%), respectively.
Page 16
Results of subgroup analyses based on gender, age range, collection site, device and clinical indication are presented in Table 6 and Figure 3 below.
Table 6: Patient-level Subgroup Analysis Results.
| Deep Capsule Prediction | Expert Reading | Sensitivity (95% CI) | Specificity (95% CI) | AUC | ||
|---|---|---|---|---|---|---|
| Lesion | Normal | |||||
| PillCam™ SB3 | Lesion | 204 | 2 | 0.981 (0.951 - 0.995) | 0.333 (0.008 - 0.906) | 0.965 (0.927 - 1.000) |
| Normal | 4 | 1 | ||||
| PillCam™ Crohn | Lesion | 16 | 0 | 0.842 (0.604 - 0.966) | 1.000 (0.025 - 1.000) | 0.842 (NA - NA) |
| Normal | 3 | 1 | ||||
| OMOM® HD | Lesion | 33 | 0 | 0.892 (0.746 - 0.970) | 1.000 (0.398 - 1.000) | 0.973 (0.914 - 1.000) |
| Normal | 4 | 4 |
Page 17
| Deep Capsule Prediction | Expert Reading | Sensitivity (95% CI) | Specificity (95% CI) | AUC | ||
|---|---|---|---|---|---|---|
| Lesion | Normal | |||||
| Male | Lesion | 112 | 1 | 0.974 (0.926, 0.995) | 0.750 (0.194 - 0.994) | 0.970 (0.933 - 1.000) |
| Normal | 3 | 3 | ||||
| Female | Lesion | 141 | 1 | 0.946 (0.897 - 0.977) | 0.750 (0.194 - 0.994) | 0.968 (0.932 - 1.000) |
| Normal | 8 | 3 | ||||
| ≤ 50 years old | Lesion | 105 | 2 | 0.991 (0.949, 1.000) | 0.500 (0.068 - 0.932) | 0.974 (0.935 - 1.000) |
| Normal | 1 | 2 | ||||
| 51-70 years old | Lesion | 87 | 0 | 0.946 (0.878 - 0.982) | 1.000 (0.158 - 1.000) | 0.973 (0.915 - 1.000) |
| Normal | 5 | 2 | ||||
| 71+ years old | Lesion | 61 | 0 | 0.924 (0.832 - 0.975) | 1.000 (0.158 - 1.000) | 0.970 (0.904 - 1.000) |
| Normal | 5 | 2 | ||||
| CHUSJ (1) | Lesion | 224 | 2 | 0.970 (0.939, 0.988) | 0.500 (0.068 - 0.932) | 0.957 (0.921 - 0.993) |
| Normal | 7 | 2 | ||||
| ManopH (2) | Lesion | 29 | 0 | 0.879 (0.718 - 0.966) | 1.000 (0.398 - 1.000) | 0.970 (0.904 - 1.000) |
| Normal | 4 | 4 | ||||
| NA | Lesion | 93 | 1 | 0.939 (0.874 - 0.972) | 0.857 (0.487 - 0.974) | 0.962 (0.925 - 1.000) |
| Normal | 6 | 6 | ||||
| Anemia | Lesion | 54 | 0 | 0.964 (0.879 - 0.990) | NA (NA - NA) | NA (NA - NA) |
| Normal | 2 | 0 | ||||
| Crohn"s Disease Diagnose | Lesion | 33 | 0 | 1.000 (0.896 - 1.000) | NA (NA - NA) | NA (NA - NA) |
| Normal | 0 | 0 | ||||
| OGIB | Lesion | 22 | 0 | 0.917 (0.742 - 0.977) | NA (NA - NA) | NA (NA - NA) |
| Normal | 2 | 0 | ||||
| Crohn"s Disease Staging | Lesion | 38 | 0 | 1.000 (0.908 - 1.000) | NA (NA - NA) | NA (NA - NA) |
| Normal | 0 | 0 | ||||
| FAP | Lesion | 3 | 1 | 1.000 (0.439 - 1.000) | 0.000 (0.000 - 0.793) | NA (NA - NA) |
| Normal | 0 | 0 | ||||
| HNPCC | Lesion | 2 | 0 | 1.000 (0.342 - 1.000) | NA (NA - NA) | NA (NA - NA) |
| Normal | 0 | 0 | ||||
| Other Polyposis | Lesion | 2 | 0 | 0.667 (0.208 - 0.939) | NA (NA - NA) | NA (NA - NA) |
| Normal | 1 | 0 | ||||
| Common | Lesion | 2 | 0 | 1.000 | NA | NA |
Page 18
| Deep Capsule Prediction | Expert Reading | Sensitivity (95% CI) | Specificity (95% CI) | AUC | ||
|---|---|---|---|---|---|---|
| Lesion | Normal | |||||
| Variable Immunodefency | Normal | 0 | 0 | (0.342 - 1.000) | (NA - NA) | (NA - NA) |
| Peutz-Jeghers Syndrome | Lesion | 3 | 0 | 1.000 (0.439 - 1.000) | NA (NA - NA) | NA (NA - NA) |
| Normal | 0 | 0 | ||||
| NET | Lesion | 1 | 0 | 1.000 (0.207 - 1.000) | NA (NA - NA) | NA (NA - NA) |
| Normal | 0 | 0 |
(1) Collection site: Centro Hospitalar Universitário São João; (2) Collection site: ManopH - Laboratório de Endoscopia e Motilidade Digestiva, Lda.
Subgroup analyses for patient-level sensitivity and specificity did not identify major differences when analyzed by gender, age, collection device, collection site and clinical indication.
Page 19
b. Image-Level Analysis
Table 7: Image-level Analysis Results.
| Deep Capsule Prediction | Expert Reading | Sensitivity (95% CI) | Specificity (95% CI) | AUC | |
|---|---|---|---|---|---|
| Lesion | Normal | ||||
| Lesion | 76492 | 1565 | 0.921 (0.919 - 0.923) | 0.880 (0.874 - 0.886) | 0.971 (0.970 - 0.972) |
| Normal | 6549 | 11476 |
Image level analysis refers to the analysis of the dataset consisting of normal (no lesion present) and abnormal (lesion present) images. This dataset may include multiple images of the same lesion from different angles. Image level sensitivity and specificity were determined to be 92.1% (95% CI: 91.9%-92.3%) and 88.0% (95% CI: 87.4%-88.6%), respectively. ROC and AUC analysis is shown below in Figure 4.
Results of subgroup analyses based on gender, age range, collection site, device and clinical indication are presented in Table 8 and Figure 5.
Page 20
Table 8: Image-level Subgroups Analysis Results.
| Deep Capsule Prediction | Expert Reading | Sensitivity (95% CI) | Specificity (95% CI) | AUC | ||
|---|---|---|---|---|---|---|
| Lesion | Normal | |||||
| PillCam™ SB3 | Lesion | 55761 | 225 | 0.912 (0.909 - 0.914) | 0.977 (0.974 - 0.980) | 0.984 (0.983 - 0.985) |
| Normal | 5407 | 9677 | ||||
| PillCam™ Crohn | Lesion | 13460 | 1338 | 0.945 (0.941 - 0.948) | 0.545 (0.527 - 0.563) | 0.934 (0.931 - 0.938) |
| Normal | 789 | 1604 | ||||
| OMOM® HD | Lesion | 7271 | 2 | 0.954 (0.949 - 0.958) | 0.990 (0.964 - 0.999) | 0.996 (0.995 - 0.998) |
| Normal | 353 | 195 | ||||
| Male | Lesion | 38701 | 98 | 0.935 (0.932 - 0.937) | 0.977 (0.971 - 0.981) | 0.990 (0.989 - 0.991) |
| Normal | 2701 | 4076 | ||||
| Female | Lesion | 37791 | 1467 | 0.908 (0.905 - 0.910) | 0.835 (0.827 - 0.842) | 0.957 (0.956 - 0.959) |
| Normal | 3848 | 7400 | ||||
| ≤ 50 years old | Lesion | 26397 | 60 | 0.898 (0.894 - 0.901) | 0.986 (0.982 - 0.989) | 0.984 (0.983 - 0.985) |
| Normal | 3008 | 4279 | ||||
| 51-70 years old | Lesion | 29756 | 163 | 0.945 (0.942 - 0.947) | 0.953 (0.945 - 0.960) | 0.985 (0.984 - 0.986) |
| Normal | 1737 | 3280 | ||||
| 71+ years old | Lesion | 20339 | 1342 | 0.919 (0.915 - 0.922) | 0.745 (0.733 - 0.757) | 0.953 (0.951 - 0.955) |
| Normal | 1804 | 3917 | ||||
| CHUSJ (1) | Lesion | 69244 | 1563 | 0.918 | 0.878 | 0.969 |
Page 21
| Deep Capsule Prediction | Expert Reading | Sensitivity (95% CI) | Specificity (95% CI) | AUC | ||
|---|---|---|---|---|---|---|
| Lesion | Normal | |||||
| Normal | 6201 | 11281 | (0.916 - 0.920) | (0.873 - 0.884) | (0.968 - 0.970) | |
| ManopH (2) | Lesion | 7248 | 2 | 0.954 (0.949 - 0.959) | 0.990 (0.964 - 0.999) | 0.996 (0.995 - 0.998) |
| Normal | 348 | 195 | ||||
| Common Variable Immunodefency | Lesion | 12499 | 0 | 0.971 (0.968 - 0.973) | NA (NA - NA) | NA (NA - NA) |
| Normal | 377 | 0 | ||||
| NA | Lesion | 27213 | 70 | 0.936 (0.934 - 0.939) | 0.988 (0.984 - 0.990) | 0.990 (0.989 - 0.991) |
| Normal | 1848 | 5572 | ||||
| OGIB | Lesion | 21898 | 27 | 0.985 (0.984 - 0.987) | 0.887 (0.840 - 0.921) | 0.989 (0.982 - 0.996) |
| Normal | 325 | 211 | ||||
| Crohn"s Disease Staging | Lesion | 3327 | 1368 | 0.985 (0.984 - 0.987) | 0.887 (0.840 - 0.921) | 0.989 (0.982 - 0.996) |
| Normal | 1434 | 2909 | ||||
| Anemia | Lesion | 8840 | 36 | 0.839 (0.832 - 0.846) | 0.980 (0.972 - 0.985) | 0.970 (0.967 - 0.973) |
| Normal | 1697 | 1744 | ||||
| Crohn"s Disease Diagnose | Lesion | 2053 | 45 | 0.779 (0.762 - 0.794) | 0.951 (0.935 - 0.963) | 0.916 (0.907 - 0.925) |
| Normal | 584 | 870 | ||||
| Peutz-Jeghers Syndrome | Lesion | 111 | 18 | 0.782 (0.707 - 0.842) | 0.878 (0.815 - 0.921) | 0.902 (0.868 - 0.936) |
| Normal | 31 | 129 | ||||
| Other Polyposis | Lesion | 223 | 0 | 0.729 (0.676 - 0.776) | NA (NA - NA) | NA (NA - NA) |
| Normal | 83 | 0 | ||||
| FAP | Lesion | 214 | 1 | 0.699 (0.646 - 0.748) | 0.976 (0.877 - 0.996) | 0.940 (0.910 - 0.970) |
| Normal | 92 | 41 | ||||
| NET | Lesion | 59 | 0 | 0.504 (0.415 - 0.593) | NA (NA - NA) | NA (NA - NA) |
| Normal | 58 | 0 | ||||
| HNPCC | Lesion | 55 | 0 | 0.733 (0.624 - 0.820) | NA (NA - NA) | NA (NA - NA) |
| Normal | 20 | 0 |
(1) Collection site: Centro Hospitalar Universitário São João; (2) Collection site: ManopH - Laboratório de Endoscopia e Motilidade Digestiva, Lda.
Subgroup analyses for image-level sensitivity and specificity did not identify major differences when analyzed by gender, age, collecting device, collecting site and clinical indication.
Page 22
Page 23
Conclusions from Lesion Detection Testing
Patient-level sensitivity was high, at 95.8%, demonstrating that the lesion detection function resulted in few false negatives for patients. Patient level specificity was 75%, suggesting that clinicians will need to carefully review Deep Capsule® findings to identify false positive predictions. At the image-level, results demonstrate sensitivity of 92.1% and specificity of 88%.
Clinical Readers and Adjudication
The clinical validation study involved 15 board-certified gastroenterologists and 5 independent expert adjudicators. Participating clinicians were certified specialists in gastroenterology and/or digestive endoscopy with a minimum of 5 years of post-fellowship clinical experience. Readers were located across the United States, Portugal, Spain, and Brazil. There was no overlap between clinical readers and adjudicators.
Retrospective Clinical Validation Study
The clinical validation dataset included 330 capsule endoscopy examinations collected between January 2021 and April 2024 from seven independent clinical centers across four countries (Portugal, Spain, Brazil, and the United States). Table 9 presents the full clinical data set demographics.
Table 9: Demographics of clinical dataset.
| Demographics Information (330 subjects) | |
|---|---|
| Mean Age, years (SD) | 58.75 (18.14) |
| <=50 years, Number of Subjects (%) | 112 (33.9%) |
| 51 - 70 years, Number of Subjects (%) | 109 (33.0%) |
| > 70 years, Number of Subjects (%) | 109 (33.0%) |
| Sex, N (%) | |
| Female | 179 (54.2%) |
| Male | 151 (45.8%) |
| Country, Number of Subjects (%) | |
| USA | 118 (35.8%) |
| Portugal | 82 (24.8%) |
| Brazil | 71 (21.5%) |
| Spain | 59 (17.9%) |
| Collection Site, Number of Subjects (%) | |
| Hospital Nove de Julho (Brazil) | 49 (14.8%) |
| Hospital Senhora da Oliveira (Portugal) | 82 (24.8%) |
| Hospital Sirio-Libanes Brasilia (Brazil) | 15 (4.5%) |
| Hospital Universitario de la Princesa (Spain) | 38 (11.5%) |
| Hospital Universitario Puerta de Hierro Majadahonda (Spain) | 21 (6.4%) |
| Hospital Vila Nova Star (Brazil) | 7 (2.1%) |
| USA Health University Hospital of South Alabama (USA) | 118 (35.8%) |
| Collection Device, Number of Subjects (%) | |
| PillCam™ SB3 | 307 (93.0%) |
| Olympus EC-10® | 21 (6.4%) |
| PillCam™ SB2 | 2 (0.6%) |
Page 24
| Indication | |
|---|---|
| Anaemia, n (%) | 166 (50.6%) |
| Crohn's disease, n (%) | 72 (22.0%) |
| Obscure gastrointestinal bleeding, n (%) | 67 (20.4%) |
| Diarrhea, n (%) | 11 (3.4%) |
| Genetic syndrome, n (%) | 4 (1.2%) |
| GI tumor, n (%) | 4 (1.2%) |
| Celiac disease, n (%) | 2 (0.6%) |
| Abdominal pain, n (%) | 1 (0.3%) |
| Weight loss, n (%) | 1 (0.3%) |
| No information | 2 |
| Race/ethnicity | |
| White or Caucasian | 252 (76.4%) |
| Black or African-american | 38 (11.5%) |
| Hispanic or Latino | 36 (10.9%) |
| Asian | 2 (0.6%) |
| American Indian or Alaska Native | 1 (0.3%) |
| Native Hawaiian or other pacific islander | 1 (0.3%) |
All clinical validation sites were independent from the institutions used for model development (Centro Hospitalar Universitário São João and ManopH, Portugal). No overlap existed between training and validation datasets.
Per-Patient Analysis
Diagnostic Yield
Table 10 summarises diagnostic yield (DY) for AI-assisted and Standard-of-Care (SoC) interpretation compared with the expert board reference standard. AI-assisted capsule endoscopy demonstrated non-inferiority to SoC (p < 0.001). Adjusted analyses accounting for prespecified covariates demonstrated consistent results.
Table 10. Diagnostic yield Per-patient level. Diagnostic yield of AI-assisted vs. standard-of-care capsule endoscopy, with comparison to expert board ground truth; raw and adjusted estimates with 95% CIs shown.
| Statistical Information | AI-aided CapE | SoC | Expert board (Ground truth) |
|---|---|---|---|
| Unadjusted (raw estimates) | |||
| n positive/ N total | 317/330 | 251/330 | 290/330 |
| Diagnostic Yield (DY) | 0.961 (0.934 - 0.977) | 0.761 (0.712 - 0.803) | 0.879 (0.839 - 0.910) |
| Difference in DY between AI-aided and SoC | 0.200 (0.149 - 0.251) Non-inferiority: established (p < 0.001) | ||
| Adjusted | |||
| Diagnostic Yield (DY) | 0.968 (0.936 - 0.984) | 0.777 (0.676 - 0.853) |
Page 25
| Difference in DY between AI-aided and SoC | 0.191 (0.083 – 0.308 ) Non-inferiority: established (p < 0.001) |
|---|
Diagnostic Yield by Subgroups
Diagnostic yield was evaluated across prespecified demographic and clinical subgroups. Performance was consistent across all evaluated categories. Table 11 presents subgroup analyses including demographic (age, gender, race/ethnicity), clinical (indication), device type, and investigation site categories.
Table 11: Diagnostic yield by demographic subgroups.
| N | AI-aided CapE | SoC | |
|---|---|---|---|
| Gender* | |||
| Female | 179 | 0.965 (0.929 - 0.983) | 0.762 (0.651 - 0.847) |
| Male | 151 | 0.970 (0.938 - 0.986) | 0.790 (0.682 - 0.868) |
| Age Range* | |||
| <=50 years | 112 | 0.967 (0.930 - 0.985) | 0.771 (0.650 - 0.859) |
| 51 - 70 years | 109 | 0.967 (0.930 - 0.985) | 0.771 (0.651 - 0.858) |
| > 71 years | 109 | 0.970 (0.936 - 0.986) | 0.788 (0.673 - 0.870) |
| Country* | |||
| USA | 118 | 0.986 (0.970 - 0.994) | 0.888 (0.823 - 0.931) |
| Portugal | 82 | 0.944 (0.894 - 0.971) | 0.651 (0.543 - 0.745) |
| Brazil | 71 | 0.963 (0.925 - 0.982) | 0.743 (0.637 - 0.827) |
| Spain | 59 | 0.973 (0.941 - 0.988) | 0.802 (0.690 - 0.880) |
| Collection device* | |||
| PillCam™ SB3 | 307 | 0.973 (0.948 - 0.986) | 0.800 (0.738 - 0.851) |
| Olympus EC-10® | 21 | 0.958 (0.891 - 0.984) | 0.719 (0.518 - 0.859) |
| Clinical indication* | |||
| Anaemia | 166 | 0.978 (0.955 - 0.990) | 0.852 (0.783 - 0.902) |
| Crohn's disease | 72 | 0.957 (0.911 - 0.980) | 0.741 (0.619 - 0.835) |
| Obscure gastrointestinal bleeding | 67 | 0.962 (0.920 - 0.982) | 0.762 (0.649 - 0.847) |
| Race/ ethnicity* | |||
| White or Caucasian | 252 | 0.968 (0.939 - 0.983) | 0.772 (0.704 - 0.829) |
| Black or African-american | 38 | 0.992 (0.975 - 0.998) | 0.935 (0.839 - 0.975) |
| Hispanic or Latino | 36 | 0.970 (0.929 - 0.988) | 0.785 (0.644 - 0.881) |
| Collection Site | |||
| USA Health University Hospital | 118 | 1.000 (0.968–1.000) | 0.788 (0.706–0.852) |
| Hospital Senhora da Oliveira | 82 | 0.878 (0.790–0.932) | 0.573 (0.465–0.675) |
| Hospital 9 de Julho | 49 | 0.980 (0.893–0.996) | 0.959 (0.863–0.989) |
| Hospital Sírio-Libanês | 15 | 1.000 (0.796–1.000) | 1.000 (0.796–1.000) |
| Hospital Univ. La Princesa | 38 | 1.000 (0.908–1.000) | 0.737 (0.580–0.850) |
| Hospital Univ. Puerta de Hierro | 21 | 0.905 (0.711–0.973) | 0.714 (0.500–0.862) |
| Hospital Vila Nova Star | 7 | 1.000 (0.646–1.000) | 0.857 (0.487–0.974) |
NOTE: analysis was not done for the device PillCam™ SB2 and for other levels of clinical indication and race/ethnicity due to very small sample sizes.
- based on a generalized linear mixed-effects models with gender and age group (fixed effects) and reader and patient (random effects); ** non-adjusted estimates obtained through logistic regression models.
Performance was consistent across evaluated subgroups. Certain categories with small sample sizes were not analyzed separately.
Per-Patient Diagnostic Performance Metrics
Page 26
Sensitivity, specificity, PPV, and NPV were calculated relative to the expert board reference standard. Table 12 summarizes sensitivity, specificity, PPV, and NPV results.
Table 12-1: Per-Patient Diagnostic Performance (Raw Estimates). Sensitivity, Specificity, Positive predictive value (PPV), and Negative predictive value (NPV) for AI-aided and SoC methods – raw estimates.
| AI-aided CapE | SoC | p | |
|---|---|---|---|
| Sensitivity | 0.972 (0.947 - 0.986) | 0.762 (0.710 - 0.807) | < 0.001 |
| Specificity | 0.125 (0.055 - 0.261) | 0.250 (0.142 - 0.402) | 0.131 |
| PPV | 0.890 (0.850 - 0.920) | 0.880 (0.835 - 0.915) | < 0.001 |
| NPV | 0.385 (0.177 - 0.645) | 0.127 (0.070 - 0.218) | - |
AI-assisted interpretation demonstrated substantially higher sensitivity compared to Standard-of-Care.
Table 12-2: Deep Capsule Model Patient-Level Classification Performance Compared to Expert Reading. Evaluation on Clinical Validation Set.
| Deep Capsule Prediction | Expert Reading | Sensitivity (95% CI) | Sensitivity (95% CI) | AUC | |
|---|---|---|---|---|---|
| Lesion | Normal | ||||
| Lesion | 317 | 13 | 1.000 (0.988 - 1.000) | 0.000 (0.000 - 0.228) | 0.816 (0.690 - 0.942) |
| Normal | 0 | 0 | |||
| Total | 317 | 13 |
Additional lesion-level analysis was conducted. The results highlight the model's prioritization of sensitivity while maintaining performance trends consistent with those observed in the internal test set.
Sensitivity and Specificity by Subgroup
Sensitivity and specificity were also evaluated across prespecified demographic, clinical, device, and site-level subgroups. AI-assisted interpretation consistently demonstrated higher sensitivity compared to Standard-of-Care. Specificity estimates varied across subgroups due to the high prevalence of positive cases in the clinical validation dataset, which limited precision of specificity estimates in certain sites. Site-level performance data are summarized below.
Page 27
Table 13: Patient-Level Sensitivity and specificity by demographic subgroups.
| # negative/ positive/ total patients | Se/Sp (95% CI): unaided (SoC) | Se/Sp (95% CI): aided (AI) | Se/Sp (95% CI): aided - unaided difference | |
|---|---|---|---|---|
| overall | 40/290/330 | Se: 76.2% (71 - 80.7) Sp: 25% (14.2 - 40.2) | Se: 97.2% (94.7 - 98.6), Sp: 12.5% (5.5 - 26.1) | Se: 21% (13.9 - 27.6) Sp: -12.5% (-34.7 - 11.9) |
| by gender | ||||
| Male | 19/132/151 | Se: 81.1% (73.5 - 86.8), Sp: 10.5% (2.9 - 31.4) | Se: 97% (92.5 - 98.8) Sp: 26.3% (11.8 - 48.8) | Se: 15.9% (5.6 - 25.3) Sp: -15.8% (-45.9 - 19.6) |
| Female | 21/158/179 | Se: 72.2% (64.7 - 78.6), Sp: 14.3% (5 - 34.6) | Se: 97.5% (93.7 - 99) Sp: 23.8% (10.6 - 45.1) | Se: 25.3% (15.1 - 34.3) Sp: -9.5% (-40.1 - 24) |
| by age range | ||||
| <=50 years | 12/100/112 | Se: 73% (63.6 - 80.7) Sp: 41.7% (19.3 - 68) | Se: 97% (91.5 - 99) Sp: 16.7% (4.7 - 44.8) | Se: 24% (10.8 - 35.4) Sp: -25% (-63.4 - 25.5) |
| 51 - 70 years | 12/97/109 | Se: 75.3% (65.8 - 82.8) Sp: 25% (8.9 - 53.2) | Se: 96.9% (91.3 - 98.9) Sp: 8.3% (1.5 - 35.4) | Se: 21.6% (8.5 - 33.1) Sp: -16.7% (-51.7 - 26.5) |
| 71+ years | 16/93/109 | Se: 80.6% (71.5 - 87.4), Sp: 12.5% (3.5 - 36) | Se: 97.8% (92.5 - 99.4), Sp: 12.5% (3.5 - 36) | Se: 17.2% (5.1 - 27.9) Sp: 0% (-32.5 - 32.5) |
| by country | ||||
| USA | 0/118/118 | Se: 78.8% (70.6 - 85.2), Sp: NA | Se: 100% (96.8 - 100) Sp: NA | Se: 21.2% (11.6 - 29.4) Sp: NA |
| Portugal | 8/74/82 | Se: 62.2% (50.8 - 72.4), Sp: 87.5% (52.9 - 97.8) | Se: 89.2% (80.1 - 94.4), Sp: 25% (7.1 - 59.1) | Se: 27% (7.7 - 43.6) Sp: -62.5% (-90.6 - 6.2) |
| Brazil | 29/42/71 | Se: 95.2% (84.2 - 98.7), Sp: 3.4% (0.6 - 17.2) | Se: 100% (91.6 - 100) Sp: 3.4% (0.6 - 17.2) | Se: 4.8% (-7.1 - 15.8) Sp: 0% (-16.6 - 16.6) |
| Spain | 3/56/59 | Se: 75% (62.3 - 84.5) Sp: 66.7% (20.8 - 93.9) | Se: 100% (93.6 - 100) Sp: 66.7% (20.8 - 93.9) | Se: 25% (9.1 - 37.7) Sp: 0% (-73.1 - 73.1) |
| by collection device | ||||
| PillCam™ SB3 | 27/270/307 | Se: 75.9% (70.5 - 80.6), Sp: 21.6% (11.4 - 37.2) | Se: 97% (94.3 - 98.5) Sp: 8.1% (2.8 - 21.3) | Se: 21.1% (13.6 - 28) Sp: -13.5% (-34.4 - 9.9) |
| Olympus EC-10® | 3/18/21 | Se: 77.8% (54.8 - 91) Sp: 66.7% (20.8 - 93.9) | Se: 100% (82.4 - 100) Sp: 66.7% (20.8 - 93.9) | Se: 22.2% (-8.6 - 45.2), Sp: 0% (-73.1 - 73.1) |
Page 28
| # negative/ positive/ total patients | Se/Sp (95% CI): unaided (SoC) | Se/Sp (95% CI): aided (AI) | Se/Sp (95% CI): aided - unaided difference | |
|---|---|---|---|---|
| PillCam™ SB2 | 0/2/2 | Se: 100% (34.2 - 100) Sp: NA | Se: 100% (34.2 - 100) Sp: NA | Se: 0% (-65.8 - 65.8) Sp: NA |
| by indication | ||||
| Abdominal pain | 0/1/1 | Se: 100% (20.7 - 100) Sp: NA | Se: 100% (20.7 - 100) Sp: NA | Se: 0% (-79.3 - 79.3) Sp: NA |
| Anaemia | 13/153/166 | Se: 80.4% (73.4 - 85.9), Sp: 30.8% (12.7 - 57.6) | Se: 97.4% (93.5 - 99) Sp: 15.4% (4.3 - 42.2) | Se: 17% (7.6 - 25.6) Sp: -15.4% (-53.3 - 29.6) |
| Celiac disease | 0/2/2 | Se: 100% (34.2 - 100) Sp: NA | Se: 100% (34.2 - 100) Sp: NA | Se: 0% (-65.8 - 65.8) Sp: NA |
| Crohn's disease | 7/65/72 | Se: 67.7% (55.6 - 77.8), Sp: 71.4% (35.9 - 91.8) | Se: 93.8% (85.2 - 97.6), Sp: 14.3% (2.6 - 51.3) | Se: 26.2% (7.4 - 42) Sp: -57.1% (-89.2 - 15.4) |
| Diarrhea | 0/11/11 | Se: 63.6% (35.4 - 84.8), Sp: NA | Se: 100% (74.1 - 100) Sp: NA | Se: 36.4% (-10.7 - 64.6) Sp: NA |
| Genetic syndrome | 0/4/4 | Se: 50% (15 - 85) Sp: NA | Se: 100% (51 - 100) Sp: NA | Se: 50% (-34 - 85) Sp: NA |
| GI tumour | 0/4/4 | Se: 75% (30.1 - 95.4) Sp: NA | Se: 100% (51 - 100) Sp: NA | Se: 25% (-44.4 - 69.9) Sp: NA |
| Obscure gastrointestinal bleeding | 20/47/67 | Se: 80.9% (67.5 - 89.6), Sp: 5% (0.9 - 23.6) | Se: 100% (92.4 - 100) Sp: 10% (2.8 - 30.1) | Se: 19.1% (2.9 - 32.5) Sp: 5% (-20.8 - 29.2) |
| Weight loss | 0/1/1 | Se: 0% (0 - 79.3) Sp: NA | Se: 100% (20.7 - 100) Sp: NA | Se: 100% (-58.7 - 100) Sp: NA |
| by race | ||||
| White or Caucasian | 32/220/252 | Se: 74.1% (67.9 - 79.4), Sp: 28.1% (15.6 - 45.4) | Se: 96.4% (93 - 98.1) Sp: 15.6% (6.9 - 31.8) | Se: 22.3% (13.6 - 30.2) Sp: -12.5% (-38.5 - 16.2) |
| Black or African-american | 0/38/38 | Se: 86.8% (72.7 - 94.2), Sp: N/A | Se: 100% (90.8 - 100) Sp: N/A | Se: 13.2% (-3.4 - 27.3) Sp: N/A |
| Hispanic or Latino | 8/28/36 | Se: 78.6% (60.5 - 89.8), Sp: 12.5% (2.2 - 47.1) | Se: 100% (87.9 - 100) Sp: 0% (0 - 32.4) | Se: 21.4% (-1.9 - 39.5) Sp: -12.5% (-47.1 - 30.2) |
| American Indian or Alaska Native | 0/1/1 | Se: 100% (20.7 - 100) Sp: N/A | Se: 100% (20.7 - 100) Sp: N/A | Se: 0% (-79.3 - 79.3) Sp: N/A |
| Asian | 0/2/2 | Se: 50% (9.5 - 90.5) Sp: N/A | Se: 100% (34.2 - 100) Sp: N/A | Se: 50% (-56.3 - 90.5) Sp: N/A |
| Native Hawaiian or other pacific islander | 0/1/1 | Se: 100% (20.7 - 100) Sp: N/A | Se: 100% (20.7 - 100) Sp: N/A | Se: 0% (-79.3 - 79.3) Sp: N/A |
| By collection site |
Page 29
| Hospital 9 de Julho (Brazil) | 23 / 26 / 49 | Se 96.2 (81.1–99.3) / Sp 4.3 (0.8–21.0) | Se 100.0 (87.1–100.0) / Sp 4.3 (0.8–21.0) | Se 3.8 (−12.2–18.9) / Sp 0.0 (−20.2–20.2) |
| Hospital Senhora da Oliveira (Portugal) | 8 / 74 / 82 | Se 62.2 (50.8–72.4) / Sp 87.5 (52.9–97.8) | Se 89.2 (80.1–94.4) / Sp 25.0 (7.1–59.1) | Se 27.0 (7.7–43.6) / Sp −62.5 (−90.6–6.2) |
| Hospital Sírio-Libanês Brasília (Brazil) | 6 / 9 / 15 | Se 100.0 (70.1–100.0) / Sp 0.0 (0.0–39.0) | Se 100.0 (70.1–100.0) / Sp 0.0 (0.0–39.0) | Se 0.0 (−29.9–29.9) / Sp 0.0 (−39.0–39.0) |
| Hospital Universitario La Princesa (Spain) | 0 / 38 / 38 | Se 73.7 (58.0–85.0) / Sp N/A | Se 100.0 (90.8–100.0) / Sp N/A | Se 26.3 (5.8–42.0) / Sp N/A |
| Hospital Universitario Puerta de Hierro (Spain) | 3 / 18 / 21 | Se 77.8 (54.8–91.0) / Sp 66.7 (20.8–93.9) | Se 100.0 (82.4–100.0) / Sp 66.7 (20.8–93.9) | Se 22.2 (−8.6–45.2) / Sp 0.0 (−73.1–73.1) |
| Hospital Vila Nova Star (Brazil) | 0 / 7 / 7 | Se 85.7 (48.7–97.4) / Sp N/A | Se 100.0 (64.6–100.0) / Sp N/A | Se 14.3 (−32.9–51.3) / Sp N/A |
| USA Health University Hospital of South Alabama (USA) | 0 / 118 / 118 | Se 78.8 (70.6–85.2) / Sp N/A | Se 100.0 (96.8–100.0) / Sp N/A | Se 21.2 (11.6–29.4) / Sp N/A |
Notes: Subgroup results with small sample sizes should be interpreted with caution due to wider confidence intervals. Specificity could not be estimated in sites with no negative patients.
Across all prespecified demographic and clinical subgroups, AI-assisted interpretation demonstrated consistently higher sensitivity compared to Standard-of-Care. Specificity estimates varied across sites and subgroups due to low prevalence of negative cases in certain centers. No clinically meaningful degradation in performance was observed across age, gender, race/ethnicity, device type, or investigation site.
Image-level analysis
Page 30
In addition to patient-level analysis, performance was evaluated at the image (frame) level. Each video frame was classified as Lesion-positive or Normal. Frame-level ground truth was established by expert reviewers blinded to AI outputs. Confusion matrices were generated comparing AI outputs and Standard-of-Care outputs against the expert reference standard.
Image-level analysis
The image-level (lesion-level) performance of Deep Capsule® was evaluated in a retrospective, multicenter clinical validation study, which included 330 capsule endoscopy examinations from seven independent clinical centers across four countries: the United States, Portugal, Brazil, and Spain.
Ground truth was established through independent manual review of full capsule endoscopy videos by two experienced gastroenterologists, with adjudication by an expert board in cases of disagreement. Reviewers establishing the reference standard were independent of the AI-assisted reading process.
Image-Level Confusion Matrix
Table 14 presents the overall image-level confusion matrix for AI-aided CapE versus the ground truth reference standard.
Table 14. Image-Level Confusion Matrix (AI-aided CapE vs Ground Truth)
| Ground Truth | Lesion | Normal | Total |
|---|---|---|---|
| AI-aided CapE = Lesion | 219,171 | 594,032 | 813,203 |
| AI-aided CapE = Normal | 14,044 | 3,348,145 | 3,362,189 |
| Total | 233,215 | 3,942,177 | 4,175,392 |
Overall Image-Level Performance and Subgroup Analyses
Page 31
Overall image-level performance metrics (sensitivity, specificity, PPV, NPV, and AUC) are presented in Table 15, including subgroup analyses across gender, age range, country, capsule device, clinical indication, and race/ethnicity.
Table 15. Image-Level Performance (AI-aided CapE vs Ground Truth) – Overall and by Subgroup (Raw Estimates)
| AI-aided CapE | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) |
|---|---|---|---|---|
| Overall | 94% (89.6 - 96.9) | 84.9% (82.6 - 87.2) | 27% (21.5 - 32.9) | 99.6% (99.3 - 99.8) |
| by gender | ||||
| female | 95.9% (92.3 - 98.3) | 84.3% (80.7 - 87.3) | 27.5% (18.6 - 36.8) | 99.7% (99.5 - 99.9) |
| male | 91.3% (82.7 - 96.9) | 85.7% (82.2 - 88.5) | 26.2% (21.1 - 31) | 99.4% (98.9 - 99.8) |
| by age range | ||||
| 0-50 | 97.6% (96.2 - 98.5) | 86.3% (82.6 - 89.7) | 28.2% (19.7 - 36.5) | 99.8% (99.8 - 99.9) |
| 51-70 | 98.4% (97.1 - 99.2) | 73.4% (65.7 - 81.4) | 21.6% (13.8 - 31.3) | 99.8% (99.7 - 99.9) |
| 71+ | 95.4% (90.5 - 98.6) | 73.9% (65.5 - 82.1) | 16.1% (9.7 - 24.2) | 99.7% (99.2 - 99.9) |
| By country | ||||
| Brazil | 96.4% (94.8 - 97.7) | 89.9% (86.3 - 92.8) | 38.7% (27.7 - 49.8) | 99.7% (99.6 - 99.9) |
| Portugal | 86.8% (70.9 - 96) | 91.9% (89.5 - 93.9) | 25.5% (15.5 - 36.3) | 99.5% (99 - 99.9) |
| Spain | 95.5% (89.1 - 98.9) | 79.3% (73.3 - 84.9) | 31.7% (18.7 - 44) | 99.4% (98.9 - 99.9) |
| USA | 97.5% (95.9 - 98.6) | 74.3% (69.8 - 79.3) | 19.6% (14.8 - 24.7) | 99.8% (99.6 - 99.9) |
| By collection device | ||||
| PillCam™ SB3 | 94.5% (89.3 - 97.4) | 85.1% (82.6 - 87.3) | 27.5% (21.6 - 34.1) | 99.6% (99.3 - 99.8) |
| PillCam™ SB2 | 59.7% (56.8 - 65.2) | 97.8% (97.4 - 98.3) | 7.2% (4.4 - 12) | 99.9% (99.8 - 99.9) |
| Olympus EC-10® | 87.2% (79.6 - 98.2) | 82.5% (75.8 - 88.5) | 21% (8.3 - 29.2) | 99.2% (97.7 - 99.9) |
| By indication | ||||
| Crohn's disease | 97.3% (94.9 - 98.6) | 88.3% (83.9 - 91.6) | 29.3% (17.5 - 41) | 99.8% (99.7 - 99.9) |
| Anemia | 92.8% (87.9 - 96.6) | 82.6% (78.9 - 86) | 20.9% (16.1 - 26.2) | 99.6% (99.2 - 99.8) |
| Obscure GI bleeding | 98.5% (95 - 99.4) | 83.8% (75.8 - 89.8) | 41% (22.9 - 57.9) | 99.8% (99.7 - 99.9) |
| Abdominal pain | 98.5% (98.5 - 98.5) | 87.4% (87.4 - 87.4) | 24.6% (24.6 - 24.6) | 99.9% (99.9 - 99.9) |
| GI tumour | 95.9% (95.5 - 98.9) | 82.5% (67.4 - 97.4) | 41.2% (9.7 - 49.2) | 99.4% (98.3 - 100) |
| Diarrhea | 57.7% (9.9 - 87.5) | 89.9% (87 - 94.9) | 20.1% (2.6 - 41.5) | 98% (96.5 - 99.9) |
| Celiac disease | 99.9% (98.6 - 100) | 84.8% (76.2 - 96.3) | 33.9% (33.8 - 34.4) | 100% (100 - 100) |
| Weight loss | 100% (100 - 100) | 88.6% (88.6 - 88.6) | 0.9% (0.9 - 0.9) | 100% (100 - 100) |
Page 32
| AI-aided CapE | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) |
|---|---|---|---|---|
| Genetic syndrome | 99.8% (98.2 - 100) | 87.1% (58.3 - 95.4) | 30.1% (7.5 - 51) | 100% (99.9 - 100) |
| By Race | ||||
| White or Caucasian | 95% (91.3 - 97.3) | 85.9% (83.5 - 88.1) | 27.6% (20.9 - 35) | 99.7% (99.5 - 99.8) |
| Black or African-american | 97.5% (93.4 - 99.3) | 70.2% (56.3 - 83.4) | 28.9% (17.8 - 37.3) | 99.6% (99 - 99.9) |
| Hispanic or Latino | 80.4% (49.9 - 98.2) | 85.8% (81.3 - 89.9) | 20.4% (10.5 - 34.1) | 99% (97.7 - 99.9) |
| American Indian or Alaska Native | 100% (100 - 100) | 59.4% (59.4 - 59.4) | 3.1% (3.1 - 3.1) | 100% (100 - 100) |
| Asian | 99% (98.9 - 100) | 86.2% (73.8 - 88.1) | 15.6% (0.6 - 42.1) | 100% (99.7 - 100) |
| Native Hawaiian or other pacific islander | 99% (98.9 - 100) | 86.2% (73.8 - 88.1) | 15.6% (0.6 - 42.1) | 100% (99.7 - 100) |
| By Collection Sites | ||||
| Hospital 9 de Julho (Brazil) | 97.5% (96 - 98.6) | 88.7% (84.1 - 92.6) | 32.6% (21 - 43) | 99.8% (99.7 - 99.9) |
| Hospital Senhora da Oliveira (Portugal) | 86.8% (70.9 - 96) | 91.9% (89.5 - 93.9) | 25.5% (15.5 - 36.3) | 99.5% (99 - 99.9) |
| Hospital Sírio-Libanês Brasília (Brazil) | 90.7% (86.4 - 94.8) | 92.9% (89.6 - 95.7) | 34.1% (16.8 - 45.3) | 99.6% (99.1 - 99.9) |
| Hospital Universitario La Princesa (Spain) | 97.2% (91.3 - 99.4) | 77.7% (68.5 - 85.2) | 35% (18.9 - 51.4) | 99.6% (99.1 - 99.9) |
| Hospital Universitario Puerta de Hierro (Spain) | 87.2% (79.6 - 98.2) | 82.5% (75.8 - 88.5) | 21% (8.3 - 29.2) | 99.2% (97.7 - 99.9) |
| Hospital Vila Nova Star (Brazil) | 96.5% (95.1 - 98.2) | 92.8% (87.2 - 97) | 73.7% (35.6 - 90.8) | 99.2% (98.4 - 99.9) |
| USA Health University Hospital of South Alabama (USA) | 97.5% (95.9 - 98.6) | 74.3% (69.8 - 79.3) | 19.6% (14.8 - 24.7) | 99.8% (99.6 - 99.9) |
At the image level, AI-aided CapE demonstrated high sensitivity (94.0%; 95% CI: 89.6–96.9) and high specificity (84.9%; 95% CI: 82.6–87.2). Sensitivity remained high across evaluated demographic and technical subgroups, and specificity estimates were generally stable across strata. No systematic degradation in overall performance was observed across major demographic or technical subgroups.
Page 33
Summary of Clinical Performance:
The clinical performance of Deep Capsule® was evaluated in a multicenter retrospective observational study designed to assess the performance of AI-assisted capsule endoscopy interpretation compared to conventional reading. The primary objective of the study was to assess the non-inferiority of AI-assisted reading for the detection of small bowel lesions using capsule endoscopy (CapE). Diagnostic yield (DY) was defined as the proportion of patients in whom at least one clinically relevant small bowel lesion was detected relative to the total number of patients examined.
Secondary objectives included the evaluation of diagnostic performance metrics (sensitivity, specificity, positive predictive value [PPV], and negative predictive value [NPV]) and comparison of mean reading time between AI-assisted and standard reading modalities.
A total of 330 capsule endoscopy examinations were included in the clinical validation dataset. Patients were enrolled from seven independent clinical centers located in Portugal, Spain, Brazil, and the United States. CapE exams were performed between January 2021 and April 2024 using three different capsule endoscopy systems: PillCam™ SB3, PillCam™ SB2, and Olympus EC-10®.
The study included 15 gastroenterologists experienced in capsule endoscopy, each with substantial prior reading experience. In the first phase of the study, videos were interpreted according to the standard of care (SoC) reading procedure by experts at the originating centers. In the second phase, anonymized videos were processed using the Deep Capsule® AI software, and AI-assisted readings were performed by independent expert readers blinded to the initial report and patient clinical data. In the AI-assisted arm, clinicians reviewed the AI-generated report and the capsule video and made the final determination regarding the presence of small bowel lesions.
An independent expert board composed of gastroenterologists with extensive capsule endoscopy experience served as the ground truth reference standard for the evaluation of diagnostic performance.
Comparison of Identification of Significant lesions vs Expert Board
| Expert Board Identification of lesions | Total | Diagnostic Yield (95% CI) | ||
|---|---|---|---|---|
| Yes | No | |||
| Expert Board (Ground Truth) | 290 | 40 | NA | 87.9% (83.9 - 91.2) |
| Standard Reading Mode (SoC) | Yes | 221 | 10 | 231 |
| No | 69 | 30 | 99 | |
| AI + Physician Reading Mode (Deep Capsule®) | Yes | 282 | 35 | 317 |
Page 34
| No | 8 | 5 | 13 |
Conclusion: AI+Physician reading resulted in non-inferior diagnostic yield compared to standard of care reading and demonstrated high agreement with the expert board. Demonstrating that the use of Deep Capsule® does not negatively impact identification of patients with small bowel pathology. In addition, mean reading time was significantly reduced with AI+Physician reading, as compared to standard reading.
All clinical performance validation was completed premarket.
Usability Evaluation
A structured usability questionnaire completed by 33 gastroenterologists demonstrated high acceptance and positive feedback. User feedback showed high usability ratings (mean scores >4.5/5 in most areas) reinforcing its perceived value in routine practice.
Pediatric Extrapolation
In this 510(k) submission, existing clinical data were not leveraged to support the use of the device in a pediatric patient population.
Post-Market Surveillance
Deep Capsule® has completed clinical performance validation, standalone algorithm testing, and software verification and validation prior to submission. No elements of device performance validation are deferred to the postmarket phase.
Deep Capsule® will be monitored through routine postmarket activities consistent with regulatory requirements applicable to Class II devices.
Labeling
The labeling for Deep Capsule® includes a detailed description of the device and the patient population for which the device is indicated for use, and instructions for use. The Instructions for Use (IFU) provide information regarding system operation, workflow integration, and interpretation of AI-assisted outputs. The labeling also includes summary
Page 35
information, including patient demographics, on the algorithm training, the non-clinical standalone performance testing and the clinical performance testing of the device in the clinical study (Study ID DC001).
The labeling includes limitations and warnings that prohibit the device from diagnosis or characterization of the lesions, and that the images and data acquired using the device are to be interpreted only by qualified medical professionals in reviewing small bowel capsule endoscopy videos. There is a warning that the device should not replace clinician decision-making. There is also a warning regarding overreliance on the device.
Conclusion:
Deep Capsule® is substantially equivalent to the legally marketed predicate device DEN230027 under 21 CFR 876.1540. Based on the technological characteristics, intended use, and clinical performance data presented in this submission, Deep Capsule® does not raise different questions of safety and effectiveness compared to the predicate device.
N/A