Search Results
Found 1 results
510(k) Data Aggregation
(273 days)
This software is a computer-assisted reading tool designed to aid endoscopists in detecting colonic mucosal lesions (such as polyps and adenomas) in real time during standard endoscopy examinations of patients undergoing screening and surveillance endoscopic mucosal evaluations. This software is used with standard White Light Imaging (WLI) and Linked Color Imaging (LCI) endoscopy imaging. This software is not intended to replace clinical decision making.
The subject device represents application of AI technology to endoscopic images to assist in detecting the presence of potential lesions. This development greatly contributes to improving the quality of colonoscopy. In recent years, computer-aided diagnosis (CAD) systems employing AI technologies have been approved and marketed as radiological medical devices for use with computed tomography (CT), X-ray, magnetic resonance imaging (MRI), and mammogram diagnostic images. In endoscopy as well, many images for diagnosis are taken. Since increasing the polyp detection rate is also in demand, CAD systems for endoscopy are being actively developed. Against this background, the company has developed this software (EW10-EC02), a new AI-based CAD system, to support Health Care Provider (HCP) detection of large intestine polyps in colonoscopic images. EW10-EC02 detects suspected large intestine polyps in the endoscope video image in real-time.
Here's a breakdown of the acceptance criteria and study findings for the EW10-EC02 Endoscopy Support Program, based on the provided document:
1. Table of Acceptance Criteria and Reported Device Performance
The document describes two main types of studies: standalone performance testing (evaluating the algorithm only) and clinical testing (evaluating human-in-the-loop performance). The acceptance criteria for the standalone performance are explicitly stated and met, while the clinical study endpoints serve as the criteria for evaluating the device's clinical benefit when assisting human readers.
Standalone Performance Acceptance Criteria & Results:
| Item | Acceptance Criteria (Implicit, based on "achieved all criteria") | Reported Performance WLI Mode | Reported Performance LCI Mode |
|---|---|---|---|
| Sensitivity per lesion (Lesion-based sensitivity) | Exceeds a defined lower limit of the 95% CI (Specific value not provided but stated as met) | 95.1% (91.1 - 98.3% CI) | 95.5% (91.5 - 98.7% CI) |
| FP Objects/Patient (Number of FPc per Case) | (Specific criteria not numerically stated, but described as "achieved all criteria") | 1.42 (1.09 - 1.81 CI) | 0.76 (0.42 - 1.21 CI) |
| Detection Persistence (Figure 1) | (Implicit: Robust correlation of detection persistence with sensitivity and FP objects/patient) | Demonstrated strong correlation | Demonstrated strong correlation |
| Frame-level performance | (Implicit: Acceptable values for TP, TN, FP, FN, sensitivity/frame, FPR/frame) | (See Table 7 for detailed values) | (See Table 7 for detailed values) |
| ROC AUC | (Implicit: High accuracy) | 0.79 (0.77-0.80 CI) | 0.87 (0.86-0.88 CI) |
| FROC Analysis | (Implicit: Supports performance) | (See Figure 4) | (See Figure 4) |
Clinical Study Endpoints & Results (serving as criteria for human-in-the-loop):
| Endpoint | Acceptance Criteria (Implicit: Superiority for APC or meeting margins for PPV; non-inferiority for FPR) | Reported Performance (CAC group vs. CC group) | P-Value / CI |
|---|---|---|---|
| Primary Endpoints | |||
| Adenoma per colonoscopy (APC) | Superiority (CAC vs. CC) | CAC: 0.990 ± 1.610; CC: 0.849 ± 1.484 | 0.018 (Superiority met) |
| Positive predictive value (PPV) | Meeting margin of -9.56% | CAC: 48.6%; CC: 54.0% | -9.56%, -1.48% (Margin met) |
| Positive percent agreement (PPA) | (Implicit: Acceptable performance) | CAC: 60.7%; CC: 66.2% | -10.50%, -2.30% |
| Secondary Endpoints of Note | |||
| Polyp per colonoscopy (PPC) | (Implicit: Acceptable performance, P-value <0.001 noted) | CAC: 1.680 ± 2.070; CC: 1.328 ± 1.791 | <0.001 (Statistically significant) |
| Serrated Lesions per Colonoscopy (SLPC) | (Implicit: Acceptable performance, P-value <0.001 noted) | CAC: 0.171 ± 0.502; CC: 0.130 ± 0.478 | 0.094 |
| Serrated Lesions Detection Rate (SLDR) | (Implicit: Acceptable performance, P-value 0.027 noted) | CAC: 13.0%; CC: 10.2% | 0.157 |
| False Positive Rate (FPR) | Non-inferiority | CAC: 17.6%; CC: 15.0% | 1.39% - 7.82% (Non-inferiority demonstrated) |
| True Histology Rate (THR) | (Implicit: Acceptable performance) | CAC: 57.0%; CC: 62.3% | -10.3%, -2.06% |
2. Sample Size Used for the Test Set and Data Provenance
Standalone Performance Testing (Algorithm only):
- Sample Size: 149 colonoscopy videos for WLI mode and 144 colonoscopy videos for LCI mode.
- WLI: 119 patients with lesions, 30 patients without lesions.
- LCI: 114 patients with lesions, 30 patients without lesions.
- Data Provenance:
- Country of origin: Implied to be primarily of Asian descent (100% Asian specified in patient demographics, Table 2). While not explicitly stated, this suggests data from an Asian country (likely Japan, given the manufacturer's location).
- Retrospective or Prospective: Retrospective, as these were "colonoscopy videos" used for evaluation, not real-time clinical use for testing.
Clinical Testing (Human-in-the-loop):
- Sample Size: 1,031 subjects analyzed (out of 1,166 enrolled).
- CAC (Computer Assisted Colonoscopy) group: 509 subjects
- CC (Conventional Colonoscopy) group: 522 subjects
- Data Provenance:
- Country of origin: United States (conducted at 12 centers in the United States).
- Retrospective or Prospective: Prospective, randomized controlled trial.
3. Number of Experts Used to Establish the Ground Truth for the Test Set and Qualifications
Standalone Performance Testing:
- The document states that "lesion detection was performed and evaluated at the lesion level (object level) in the actual clinical cases". However, it does not specify the number of experts used to establish the ground truth for the standalone test set. It also does not detail their qualifications (e.g., years of experience, board certification). The data originated from "actual clinical cases," implying that the ground truth was derived from the in-vivo findings and subsequent histopathology from those cases.
Clinical Testing:
- The ground truth for this study was established by histopathology of resected polyps. This is considered the gold standard for lesion characterization. Endoscopists performed the resections during the colonoscopies. The document does not specify the number of pathologists or their qualifications who reviewed the specimens, but typically, this is done by qualified histopathologists.
4. Adjudication Method for the Test Set
Standalone Performance Testing:
- The document does not explicitly describe an adjudication method for the ground truth of the standalone test set. It states that sensitivity per lesion was the primary endpoint and refers to evaluation values calculated per frame and per case, based on detection persistence. This suggests a direct comparison to the established ground truth without a specific expert adjudication process for discrepancies.
Clinical Testing:
- The primary ground truth for the clinical study was histopathology. For endpoints like APC, PPV, and PPA, histopathology results directly determined the presence and type of lesions. There is no mention of an explicitly described adjudication method (like 2+1 or 3+1 expert review) for the clinical study results, as the histopathological diagnosis serves as the definitive ground truth for the resected tissue.
5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was done, and the Effect Size of Improvement with AI vs. without AI Assistance
- A MRMC study was done for the clinical performance evaluation. This was a multi-center, prospective, randomized controlled trial comparing Computer Assisted Colonoscopy (CAC group) with AI assistance to Conventional Colonoscopy (CC group) without AI assistance.
- Effect Size of Improvement (How human readers improve with AI vs. without AI assistance):
- Adenoma Per Colonoscopy (APC): The mean number of adenomas per colonoscopy increased from 0.849 in the Conventional Colonoscopy (CC) group to 0.990 in the Computer Assisted Colonoscopy (CAC) group. This represents an absolute difference of 0.141 (95% CI: 0.01, 0.28), which was statistically significant (p=0.018). This demonstrates that AI assistance led to an increase in the detection of adenomas per patient.
- Polyp Per Colonoscopy (PPC): The mean number of polyps per colonoscopy increased from 1.328 in the CC group to 1.680 in the CAC group. This was also statistically significant (p<0.001).
- Serrated Lesions Per Colonoscopy (SLPC): While not statistically significant at 0.05, the mean SLPC increased from 0.130 in CC to 0.171 in CAC (p=0.094).
- Serrated Lesions Detection Rate (SLDR): The rate increased from 29.1% in CC to 35.6% in CAC, with a p-value of 0.027, indicating a statistically significant improvement in the detection rate of serrated lesions.
6. If a Standalone (Algorithm Only without Human-in-the-Loop Performance) was done
- Yes, a standalone performance testing was done. This is detailed under the "Non-clinical Performance Testing" section. The evaluation assessed the device's object-level, frame-level, and overall algorithmic performance.
7. The Type of Ground Truth Used
Standalone Performance Testing:
- The ground truth was established by annotations of lesions identified in the colonoscopy videos, with the ultimate confirmation likely tied to the patient's pathology results for those lesions. It refers to "actual clinical cases" where "lesion detection was performed and evaluated at the lesion level (object level)".
Clinical Testing:
- The definitive ground truth for the clinical study was histopathology (pathology results) of resected polyps. This is explicitly stated for the primary endpoints (APC, PPV, PPA and many secondary endpoints like ADR, SLPC, etc.) which were based on "histologically confirmed" findings.
8. The Sample Size for the Training Set
- The document does not explicitly state the sample size used for the training set of the AI algorithm. It only discusses the dataset used for standalone performance testing (test set) and the clinical study population.
9. How the Ground Truth for the Training Set was Established
- The document does not explicitly describe how the ground truth for the training set was established. It mentions the EW10-EC02 utilizes an "artificial intelligence-based algorithm to perform the polyp detection function." Typically, for such AI systems, the training data is extensively annotated by medical experts (e.g., endoscopists, pathologists) to provide the ground truth for the algorithm to learn from. However, the specific process (e.g., number of annotators, their qualifications, adjudication methods for training data) is not detailed in this submission.
Ask a specific question about this device
Page 1 of 1