Search Results

Automatic Anatomy Recognition (AAR) is a software-only medical device intended for use by technicians and trained physicians to derive contours of anatomical structures from computed tomography studies for input to a radiation treatment planning system. It is only intended to work for anatomical structures in the head & neck and thoracic body regions. It is not for use on patients below 18 years of age and it relies on third party treatment planning systems to display and edit the contours.

Automatic Anatomy Recognition product for radiation therapy planning (AAR) is a software-only medical device and is deployed on a cloud-based platform. AAR is intended to be used on adults undergoing treatment that requires the identification of anatomical structures in the body considered to be "organs at risk" (OAR). AAR is intended to be used in the head and thoracic body regions. AAR automatically processes computed tomography (CT) studies and produces contours with no human intervention. AAR does not provide the capability to modify contours. If adjustments are required, they must be performed on another system.

Here's a breakdown of the acceptance criteria and study information based on the provided text, focusing on the Automatic Anatomy Recognition (AAR) device.

1. Table of Acceptance Criteria and Reported Device Performance

The document mentions "segmentation accuracy non-inferiority using DICE similarity coefficients" and "mean 95% Hausdorff Distance (HD)" calculations as performance testing methods. However, it does not explicitly state specific numerical acceptance criteria (e.g., "DICE score > 0.8" or "HD < 2mm") for each anatomical structure. It only indicates that these metrics were used to evaluate performance against the predicate.

Therefore, the table below reflects what is stated in the document and highlights the missing specific numerical criteria.

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

• Test Set Sample Size: Not explicitly stated in the provided text. The document mentions "segmentation performance tests" and "additional tests," but no specific number of cases or images for the test set is given.
• Data Provenance: Not explicitly stated. The document does not mention the country of origin of the data or whether it was retrospective or prospective.

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

• Number of Experts: Not explicitly stated.
• Qualifications of Experts: Not explicitly stated. The document refers to "ground truth" (implicitly through DICE and HD calculations which require a reference standard), but does not detail how this ground truth was established or by whom.

4. Adjudication Method for the Test Set

• Adjudication Method: Not explicitly stated. Given that the number of experts and their roles are not detailed, the adjudication method (e.g., 2+1, 3+1, none) for establishing ground truth is not described in this document.

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

• MRMC Study Done: No, the document does not indicate that a Multi-Reader Multi-Case (MRMC) comparative effectiveness study was done. The study described focuses on standalone algorithm performance comparison against a predicate device's method using quantitative metrics. Human readers' improvement with or without AI assistance is not discussed.

6. Standalone (Algorithm Only) Performance Study

• Standalone Study Done: Yes. The document refers to "automated segmentation accuracy" and states that "AAR automatically processes computed tomography (CT) studies and produces contours with no human intervention." This confirms that a standalone algorithm performance without human-in-the-loop was performed. The evaluation used DICE similarity coefficients and Mean 95% Hausdorff Distance (HD) calculations.

7. Type of Ground Truth Used

• Type of Ground Truth: Expert consensus (implied). While not explicitly stated as "expert consensus," the use of metrics like DICE similarity coefficient and Hausdorff Distance for evaluating segmentation accuracy heavily implies the existence of a meticulously hand-segmented gold standard, which is typically created by medical experts (e.g., radiologists, radiation oncologists, or dosimetrists) who define the "true" boundaries of anatomical structures. The term "ground truth" itself is used in the context of performance testing, and for contouring, this usually refers to expert-drawn contours.

8. Sample Size for the Training Set

• Training Set Sample Size: Not explicitly stated. The document mentions the use of "Deep Learning" and that the algorithm is "trained with clinical and/or artificial radiological images," but provides no details on the size of the training dataset.

9. How Ground Truth for the Training Set Was Established

• Ground Truth for Training Set: Not explicitly stated. Given that the device uses "Deep Learning contouring," it necessitates a large dataset of images with corresponding accurate ground truth contours for training. However, the document does not detail the specific process or individuals involved in establishing these ground truth contours for the training data.

Ask a specific question about this device