(222 days)
This device is indicated to acquire and display cross-sectional volumes of the whole the head, with the capability to image whole organs in a single rotation. Whole organs include, but are not limited to brain, heart, pancreas, etc.
The Aquilion ONE has the capability to provide volume sets of the entire organ. These volume sets can be used to perform specialized studies, using indicated software, of the whole organ by a trained and qualified physician.
FIRST is an iterative reconstruction algorithm intended to reduce exposure dose and improve high contrast spatial resolution for abdomen, pelvis, chest, cardiac, extremities and head applications.
AiCE is a noise reduction algorithm that improves image quality and reduces image noise by employing Deep Convolutional Network methods for abdomen, pelvis, lung and cardiac applications.
Aguilion ONE (TSX-305A/6) V8.9 with AiCE is a whole body multi-slice helical CT scanner, consisting of a gantry, couch and a console used for data processing and display. This device captures cross sectional volume data sets used to perform specialized studies, using indicated software/hardware, by a trained and qualified physician. This system is based upon the technology and materials of previously marketed Canon CT systems. In addition, the subject device incorporates the latest reconstruction technology, AiCE (Advanced intelligent Clear-IQ Engine), intended to reduce image noise and improve image quality by utilizing Deep Convolutional Neural Network methods. These methods can more fully explore the statistical properties of the signal and noise. By learning to differentiate structure from noise, the algorithm produces fast, high quality CT reconstruction.
Acceptance Criteria and Study Proving Device Performance: Canon Medical Systems Corporation Aquilion ONE (TSX-305A/6) V8.9 with AiCE
This document outlines the acceptance criteria and the study conducted to prove that the Aquilion ONE (TSX-305A/6) V8.9 with AiCE (Advanced intelligent Clear-IQ Engine) CT system meets its performance claims, as detailed in the provided FDA 510(k) summary (K183046).
The AiCE algorithm is a noise reduction algorithm that utilizes Deep Convolutional Neural Network methods to improve image quality and reduce image noise for abdomen, pelvis, lung, and cardiac applications. The primary goal of the study was to demonstrate that the AiCE system provides improved image quality, reduced noise, and better low-contrast detectability compared to the predicate device and other reconstruction methods, while maintaining diagnostic quality.
1. Table of Acceptance Criteria and Reported Device Performance
The acceptance criteria for the Aquilion ONE (TSX-305A/6) V8.9 with AiCE are based on the comparison of its image quality metrics against established benchmarks, primarily the predicate device (Aquilion ONE (TSX-305A/3) V8.3 with FIRST 2.1) and other reconstruction methods (AIDR 3D, Filtered Back Projection).
| Metric / Claim Category | Acceptance Criteria (Implicit/Explicit) | Reported Device Performance (with AiCE) |
|---|---|---|
| Image Quality (General) | Images must be of diagnostic quality for intended applications (abdomen, pelvis, lung, cardiac). | Representative abdomen/pelvis, lung, and cardiac diagnostic images, reviewed by an American Board Certified Radiologist, were obtained and confirmed to be of diagnostic quality with AiCE reconstruction. |
| Quantitative Spatial Resolution | Improved quantitative spatial resolution over AIDR 3D. | Improvement claim of 7.4 lp/cm at 10% of the MTF for AiCE, relative to 5.7 lp/cm at 10% of the MTF when using AIDR 3D/STANDARD. This represents a significant improvement. |
| Low Contrast Detectability (LCD) | Improved low-contrast detectability over AIDR 3D. | |
| Demonstrate specific LCD performance for routine scanning. | A 12.2% improvement in low contrast detectability compared to AIDR 3D (at the same dose). | |
| Demonstrated low-contrast detectability of 1.5 mm at 0.3% contrast with 22.6mGy. | ||
| Noise Reduction | Improved quantitative noise reduction over AIDR 3D. | A 29.2% noise reduction (at the same dose) compared to AIDR 3D. |
| Dose Reduction | Demonstrate significant dose reduction capabilities. | A dose reduction of 79.6-82.4% compared to filtered back projection (for body applications, based on model observer evaluation). |
| Noise Texture/Appearance | Noise appearance/texture produced by AiCE should be similar to or better than standard filtered backprojection, and less artificial than other iterative reconstructions. | A phantom study determined that the noise appearance/texture produced by the AiCE reconstruction is more similar to standard filtered backprojection than texture produced by the FIRST reconstruction. This implies a more desirable, less "plastic" appearance often associated with strong iterative reconstructions. |
| General CT Image Metrics | Performance must be substantially equivalent to or better than the predicate device across various standard CT image quality metrics: Contrast-to-Noise Ratios (CNR), CT Number Accuracy, Uniformity, Slice Sensitivity Profile (SSP), Modulation Transfer Function (MTF)-Wire, Modulation Transfer Function (MTF)-Edge, Standard Deviation of Noise (SD), Noise Power Spectra (NPS), and Pediatric water phantom. | AiCE was demonstrated to be "substantially equivalent to the predicate device as demonstrated by the results of the above testing" across these metrics. While specific numbers are not provided for each, the overall statement indicates compliance with expected performance standards for a CT system, with the specific improvements highlighted above adding to this foundational equivalence. |
2. Sample Size Used for the Test Set and Data Provenance
- Sample Size for Test Set: The documentation does not explicitly state a specific number of cases or scans used for the quantitative phantom studies or the qualitative diagnostic image review. The qualitative evaluation mentions "Representative abdomen/pelvis, lung, and cardiac diagnostic images," implying a selection of cases rather than a large cohort. The quantitative evaluations used various phantoms.
- Data Provenance: The studies were phantom-based for quantitative metrics and included "Representative ... diagnostic images" for qualitative assessment. There is no information provided regarding the country of origin of the diagnostic images, nor whether they were retrospective or prospective patient data. Given the context of a 510(k) submission primarily relying on technical performance and equivalence, large-scale patient outcome studies (prospective or retrospective) are often not required if technical equivalence and safety can be demonstrated through phantom studies and limited clinical image review.
3. Number of Experts Used to Establish Ground Truth for the Test Set and Qualifications
- Number of Experts: For the qualitative assessment of diagnostic images, it is stated that images were "reviewed by an American Board Certified Radiologist." This indicates one expert was used for this part of the evaluation.
- Qualifications of Experts: The expert was an "American Board Certified Radiologist." No specific years of experience are mentioned. For the quantitative phantom studies, experts establishing ground truth would typically be physicists or engineers, but this is not explicitly detailed.
4. Adjudication Method for the Test Set
For the qualitative assessment of diagnostic images, the documentation states that images were "reviewed by an American Board Certified Radiologist, and it was confirmed that the AiCE reconstructed images were of diagnostic quality." This implies a single-reader assessment for confirming diagnostic quality, rather than a multi-reader or consensus-based adjudication method (e.g., 2+1 or 3+1). For the phantom studies, adjudication methods for ground truth are generally not applicable as the "ground truth" is derived from the known properties of the phantoms or precise measurements.
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
No, an MRMC comparative effectiveness study was not explicitly reported in this 510(k) summary. The evaluation of the AiCE algorithm appears to have focused on:
- Quantitative measurements using phantoms (spatial resolution, LCD, noise, dose reduction).
- Qualitative review by a single radiologist to confirm diagnostic quality of "representative" patient images.
- A phantom-based "noise texture reader study" to compare noise appearance, but this is not described as a full MRMC study for diagnostic accuracy or reader performance.
Therefore, no effect size or improvement in human reader performance with AI assistance vs. without AI assistance is detailed in this submission. The AiCE algorithm acts as an inherent image reconstruction/processing component of the CT system, improving the fundamental image quality before interpretation, rather than an AI-assisted diagnostic tool for specific pathologies that would typically undergo MRMC studies.
6. Standalone (Algorithm Only Without Human-in-the-Loop Performance) Study
Yes, a standalone (algorithm only) performance assessment was conducted for the key imaging metrics. The phantom studies directly evaluate the performance of the AiCE algorithm in terms of:
- Quantitative Spatial Resolution (7.4 lp/cm MTF)
- Quantitative Body LCD Improvement (12.2% better)
- Quantitative Noise Reduction (29.2% less noise)
- Quantitative Dose Reduction (79.6-82.4% reduction vs. FBP)
- Noise Texture/Appearance (more similar to FBP than FIRST)
- General CT Image Quality metrics (CNR, CT Number Accuracy, Uniformity, SSP, MTF, SD, NPS, Pediatric water phantom).
These measurements assess the intrinsic performance of the AiCE algorithm's output (the reconstructed image) independent of human interpretation.
7. Type of Ground Truth Used
- Quantitative Studies (Spatial Resolution, LCD, Noise, Dose): The ground truth was established through phantom studies. Phantoms have precisely known physical properties and are designed to provide a measurable "truth" for imaging performance metrics (e.g., specific object sizes for spatial resolution, known contrast differences for LCD).
- Qualitative Assessment (Diagnostic Image Quality): The ground truth was expert consensus / opinion from a "single American Board Certified Radiologist" who confirmed the "diagnostic quality" of the AiCE reconstructed images.
- Noise Texture Reader Study: The ground truth for noise texture comparison was based on the collective assessment/preference of readers in that dedicated phantom study, establishing what "more similar to standard filtered backprojection" means.
8. Sample Size for the Training Set
The 510(k) summary does not provide details on the sample size used for training the AiCE Deep Convolutional Neural Network. This information is typically proprietary and not always required in detail for a 510(k) submission, especially when the device is an image reconstruction algorithm rather than a diagnostic AI that provides a specific clinical output (e.g., detection of a disease). The focus of this submission is on the output performance of the trained algorithm in terms of image quality metrics.
9. How the Ground Truth for the Training Set Was Established
The 510(k) summary does not provide details on how the ground truth was established for the training set of the AiCE Deep Convolutional Neural Network. For deep learning image reconstruction algorithms, the training process often involves:
- Pairs of "noisy" and "clean" image data: The network learns to transform low-dose/noisy inputs into high-quality outputs. The "clean" ground truth might be derived from high-dose scans, or synthetic data generated with known characteristics, or even "ideal" images produced by traditional, more time-consuming reconstruction methods.
- Simulations: Using physics-based models to simulate noise and artifacts, then generating ideal "ground truth" images.
- Existing clinical data: Using a large dataset of patient images, processed or curated to serve as ground truth for noise reduction and image enhancement.
Without explicit information, the specific method used to establish ground truth for AiCE's training set remains undetailed in this FDA submission.
§ 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.