Search Results

The SKOUT system is a software device designed to detect potential colorectal polyps in real time during colonoscopy examinations. It is indicated as a computer-aided detection tool providing colorectal polyps location information to assist qualified and trained gastroenterologists in identifying potential colorectal polyps during colonoscopy examinations in adult patients undergoing colorectal cancer screening or surveillance.

The SKOUT system is only intended to assist the gastroenterologist in identifying suspected colorectal polyps and the gastroenterologist is responsible for reviewing SKOUT suspected polyp areas and confirming the presence of a polyp based on their own medical judgment. SKOUT is not intended to replace a full patient evaluation, nor is it intended to be relied upon to make a primary interpretation of endoscopic procedures, medical diagnosis, or recommendations of treatment/course of action for patients. SKOUT is indicated for white light colonoscopy only.

The SKOUT™ system is a software-based computer aided detection (CADe) system for the analysis of high-definition endoscopic video during colonoscopy procedures. The SKOUT™ system is intended to aid gastroenterologists with the detection of potential colorectal polyps during colonoscopy by providing an informational visual aid on the endoscopic monitor using trained software that processes the endoscopic video in real time.

Users will primarily interact with the SKOUT™ system by observing the software display, including the polyp detection box and device status indicator signal.

Acceptance Criteria and Device Performance for SKOUT™ System

1. Table of Acceptance Criteria and Reported Device Performance

The provided document doesn't explicitly state "acceptance criteria" in a separate section. However, it presents the results of "Non-clinical performance testing" and "Clinical Testing" with specific metrics and confidence intervals, which serve as the performance benchmarks the device has met.

Based on the information provided, here's a table summarizing the implicit acceptance criteria (as demonstrated by the positive study outcomes) and the reported device performance:

2. Sample Sizes and Data Provenance

• Standalone Algorithm Performance Testing:

  • Test Set Sample Size: 79 HD videos containing 94 polyps.
  • Data Provenance: Not specified regarding country of origin. The document states "a predefined set of 79 HD videos". It's implied to be retrospective as it's a "set of videos" used for evaluation.

• Clinical Testing (MCRCT):

  • Test Set Sample Size: A total of 1,359 patients were included in the modified Intention To Treat (mITT) population.
    • AI-aided arm (SKOUT™ system): 682 patients
    • Control arm (Standard colonoscopy): 677 patients
  • Data Provenance: Multi-center, prospective, randomized controlled trial conducted in the United States.

3. Number of Experts and Qualifications for Ground Truth Establishment

• Standalone Algorithm Performance Testing:

  • Number of Experts: "Expert gastroenterologists" (plural, but specific number not provided).
  • Qualifications: "Expert gastroenterologists" who reviewed and either validated, rejected, or created new labels. No specific details like years of experience are given, but "expert" implies significant experience.

• Clinical Testing:

  • Number of Experts: The endoscopists participating in the trial served as the "human experts" whose findings formed the basis of the clinical ground truth. The eligibility requirements for these providers were: a) United States board-certified gastroenterologist, b) Has performed at least 1,000 colonoscopies, and c) Has an ADR (Adenoma Detection Rate) greater than or equal to 25%.
  • Qualifications: Board-certified gastroenterologists with extensive experience (minimum 1,000 colonoscopies and a good ADR).

4. Adjudication Method for the Test Set

• Standalone Algorithm Performance Testing: "Ground truth was defined as data reviewed and either validated or created by expert gastroenterologists through a process referred to as gastroenterologist review. During gastroenterologist review, experts reviewed and either validated, rejected new labels post primary annotation." This implies a consensus process, but the exact number of experts involved in each review/adjudication step (e.g., 2+1, 3+1) is not explicitly stated. It mentions a "team of trained annotators" initially, then "expert gastroenterologists" for validation/adjudication.

• Clinical Testing: For the clinical trial, the ground truth was based on the findings confirmed by the gastroenterologist during the colonoscopy and subsequent pathology findings. As this was a clinical trial involving real-time diagnosis and procedure (polyp removal), standard clinical practice implicitly serves as the "adjudication" for polyp presence, with pathology confirming the nature of the resected tissue. There wasn't a separate, explicit "adjudication method" in the sense of multiple independent readers reviewing the same data retrospectively; rather, it was a prospective clinical study using the treating physician's findings as the clinical gold standard, confirmed by pathology when resected.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study

• Yes, a clinical study was performed that can be considered a form of comparative effectiveness study, although not explicitly labeled as an MRMC study in the traditional sense of multiple readers evaluating the same cases. It was a randomized controlled trial comparing two groups of patients (and thus, two groups of cases and different physicians): one with AI assistance and one without.
• Effect Size of Human Readers Improvement with AI vs. without AI assistance:
  • The primary performance endpoint was Adenomas Per Colonoscopy (APC).
  • Without AI (Control Arm): Mean APC = 0.830
  • With AI Assistance (Treatment Arm): Mean APC = 1.054
  • Effect Size / Improvement: The difference in APC was +0.224 (Treatment - Control), with a p-value of 0.002. This indicates a statistically significant increase in the number of adenomas detected per colonoscopy when the SKOUT™ system was used. This is a direct measure of how much human readers (gastroenterologists) improved their adenoma detection rate when assisted by the AI.

6. Standalone (Algorithm Only) Performance Study

• Yes, a standalone algorithm performance testing was done.
• Details: "The Standalone Performance Assessment was performed to assess trends in the detection and classification of the SKOUT™ system and determine its ability to discriminate between normal mucosa and polyp tissue." This included analysis of 79 HD videos. Performance metrics such as Object Level TPR, Object Level FP, Frame Level TPR, and Frame Level FPR were evaluated.

7. Type of Ground Truth Used

• Standalone Algorithm Performance Testing:

  • "Gastroenterologist review" followed by pathology findings for Object Level TPR (polyps confirmed by pathology findings).
  • For Frame Level metrics, the ground truth for polyp detection frames was established by expert gastroenterologists, with "normal mucosa" as the negative ground truth.

• Clinical Testing:

  • Clinical Diagnosis from Trained Gastroenterologist + Pathology: For the APC endpoint, adenomas were confirmed through histopathology after resection. For PPA/PPV, the ground truth was based on resected polyps confirmed by pathology.

8. Sample Size for the Training Set

• The document does not explicitly state the sample size used for the training set of the SKOUT™ system's AI algorithm. It only details the test set for standalone performance and the patient cohort for the clinical trial.

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

• The document does not explicitly detail how the ground truth for the training set was established. It mentions that the SKOUT™ system "utilizes an artificial intelligence-based algorithm to perform the polyp detection function" and refers to "trained software," implying a training phase. However, the specific process for ground truth establishment for the training data is not provided in this document, only for the evaluation (test) datasets.

Ask a specific question about this device