(305 days)
The Precise Image is a reconstruction software application for a Computed Tomography X-Ray System intended to produce images of the head and body by computer reconstruction of x-ray transmission data taken at different angles and planes. These devices may include signal analysis and display equipment supports, components and accessories. Precise Image has been evaluated and available on preselected reference protocols for adult subjects. Precise Image is not indicated for use in pediatric subjects.
The CT system with Precise Image is indicated for head, whole body and vascular X-ray Computed Tomography applications. These scanners are intended to be used for diagnostic imaging.
Precise Image uses an Artificial Intelligence powered reconstruction that is designed for low radiation dose, provides lower noise, and improves low contrast detectability.
The proposed Precise Image is a reconstruction software application that may be used on a Philips whole-body computed tomography (CT) X-Ray System. Precise Image is a robust reconstruction software application, utilizing technological advancements in Artificial Intelligence and a Convolutional Neural Networks (CNN), When used, Precise Image generates CT images that provides an image appearance similar to traditional FBP images while reducing dose and improving image quality.
The implemented algorithm includes 5 user-adjustable settings to match the Radiologist's preference for dose reduction and image quality.
The proposed Precise Image reconstruction has been trained on and may be used on the currently marketed predicate device Philips Incisive CT System (K180015).
Here's a breakdown of the acceptance criteria and the study details for the Philips Precise Image device, based on the provided text:
Acceptance Criteria and Device Performance
| Acceptance Criteria | Reported Device Performance |
|---|---|
| Low Contrast Resolution (20 cm Catphan phantom) | 5 mm @ 0.3% @ 5.5 mGy CTDIvol (Improved, better low contrast resolution at lower dose levels compared to predicate's 4 mm @ 0.3% @ 22 mGy CTDIvol) |
| Noise Reduction and Low Contrast Detectability | Achieving up to 85% lower noise at 80% lower dose and 60% better low contrast detectability (Improved compared to standard mode, which is the baseline for the claim) |
| Noise Power Spectrum (NPS) Shift | Where noise is reduced by at least 50%, the system shall shift the noise power spectrum of images by no more than 6% as compared to the same data reconstructed without Precise Image. (Will not shift NPS more than 6%) |
| Application | Head, Body, and Vascular (Matches predicate's Head, Body, Vascular, and Cardiac applications in relevant scan types) |
| Scan Regime | Continuous Rotation (Identical to predicate) |
| Scan Field of View | Up to 500 mm (Identical to predicate) |
| Minimum Scan Time | 0.35 sec for 360° rotation (Identical to predicate) |
| Noise in Standard Mode (21.6 cm water-equivalent) | 0.27% at 27 mGY (Identical to predicate) |
| Compliance with Standards and Guidance | Maintains compliance with IEC 60601-1, IEC 60601-1-2, IEC 60601-1-3, IEC 60601-1-6, IEC 60601-2-44, IEC 62304, ISO 10993-1, ISO 14971, Guidance for Industry and FDA Staff – Guidance for the Content of Premarket Submissions for Software Contained in Medical Devices, Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Discussion Paper. |
| Image Quality (Clinical Evaluation) | All images were evaluated to have good image quality by certified radiologists. |
| Diagnostic Confidence, Sharpness, Noise Level, Image Texture, and Artifacts | Evaluated on a five-point Likert scale, demonstrating substantial equivalence to the predicate. |
Study Details
-
Sample Size for Test Set and Data Provenance:
- Sample Size: 55 image set pairs.
- Data Provenance: The document states "Sample clinical images are provided with this submission," implying these are real clinical images. No specific country of origin is mentioned, nor is it explicitly stated if the data is retrospective or prospective. However, given they are "clinical images" and used for evaluation, it's highly likely they are retrospective images from existing clinical practice.
-
Number of Experts and Qualifications:
- Number of Experts: 6 board-certified radiologists.
- Qualifications: "board certified radiologists." No specific years of experience or subspecialty are provided.
-
Adjudication Method for the Test Set:
- The document implies individual evaluations by each of the 6 radiologists on a Likert scale for various image attributes. It does not mention any explicit adjudication method (like 2+1 or 3+1 consensus) for the ground truth of the test set itself. The radiologists assessed "Diagnostic Confidence. Sharpness, Noise level. Image texture and Artifacts." The study compares the proposed device images against predicate device images, with the radiologists providing their individual assessment on these attributes.
-
Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study:
- Was it done? Yes, a comparative image evaluation study was performed by 6 board-certified radiologists on 55 image set pairs.
- Effect size of human readers improvement with AI vs without AI assistance: The document states that the evaluation was to "evaluate Diagnostic Confidence. Sharpness, Noise level. Image texture and Artifacts on a five point Likert scale" demonstrating "substantial equivalence to the currently marketed predicate device Philips Incisive CT (K180015)." It highlights improvements in low contrast resolution, noise reduction, and low contrast detectability of the device itself compared to the predicate/standard mode, but does not quantify human reader improvement (e.g., AUC, sensitivity, specificity) with AI assistance versus without it. The evaluation focused on image quality and characteristics, not diagnostic accuracy changes for the human reader.
-
Standalone (Algorithm Only) Performance Study:
- Yes, the performance characteristics like "Low Contrast Resolution," "Noise Reduction and Low Contrast Detectability," and "Noise Power Spectrum" are measurements of the algorithm's output (the reconstructed image) and are done in a standalone manner without human intervention influencing these specific metrics. The clinical image evaluation by radiologists also assesses the output of the algorithm relative to the predicate.
-
Type of Ground Truth Used:
- For the quantitative technical specifications (e.g., low contrast resolution, noise, NPS), the ground truth is based on phantom measurements (e.g., "20 cm Catphan phantom," "21.6 cm water-equivalent").
- For the clinical image evaluation, the "ground truth" for comparison is the predicate device's images (Incisive CT and Brilliance iCT), with radiologists evaluating the attributes of the Precise Image compared to these established images. There is no mention of a separate, definitive, clinical ground truth (e.g., pathology, clinical outcomes) for the diagnosis from these images. The radiologists are evaluating image quality characteristics and comparing them.
-
Sample Size for the Training Set:
- Not explicitly stated in the provided text. The document mentions, "The proposed Precise Image reconstruction has been trained on and may be used on the currently marketed predicate device Philips Incisive CT System (K180015)." However, it doesn't give a specific number of images or cases used for training.
-
How the Ground Truth for the Training Set was Established:
- Not explicitly stated in the provided text. It mentions the device "has been trained on" the predicate device's data, implying that the established high-quality images from the predicate device likely served as a reference or ground truth for the AI training process to guide the AI in producing similar or improved image characteristics. However, the specific method of ground truth establishment for training data is not detailed.
§ 892.1750 Computed tomography x-ray system.
(a)
Identification. A computed tomography x-ray system is a diagnostic x-ray system intended to produce cross-sectional images of the body by computer reconstruction of x-ray transmission data from the same axial plane taken at different angles. This generic type of device may include signal analysis and display equipment, patient and equipment supports, component parts, and accessories.(b)
Classification. Class II.