(40 days)
The Low Dose CT Lung Cancer Screening Option for the SCENARIA and SUPRIA CT systems is indicated for using low dose CT for lung cancer screening. The screening must be conducted with the established protocols that have been approved and published by a governmental body, a professional medical society, and/or Hitachi.
There are not any functional, performance, feature, or design changes required for the CT systems which the option is applied:
- SCENARIA Phase 3 Whole-Body X-ray CT System K150595
- SUPRIA Whole-body X-ray CT System Phase 3 K163528
Because neither of the CTs will require hardware or software modifications the subject device will include: - Three reference LCS protocols (small, average, large patient) for each CT System
- Protocols will be loaded onto the system, there will be no need for installation instructions
- Low Dose CT Lung Cancer Screening Option instruction manual
The reconstruction method for the LCS protocols is Filtered Back Projection with no iterative reconstruction method. The reconstruction algorithm used was a 21 Lung which is common to demonstrate lung tissues nodules and other lung pathology.
In addition, Beam Hardening Correction is utilized in the reconstruction process. The beam hardening correction applied to the lung reconstruction algorithm corrects image quality degradation due to radiation hardening caused by metals and other dense subject matter such as shoulders, etc. Hitachi does not apply any other tools or software in the reconstruction process.
The provided text describes the substantial equivalence determination for the "Low Dose CT Lung Cancer Screening Option" from Hitachi Healthcare Americas. The submission focuses on demonstrating that the new option, which consists of reference LCS protocols for existing SCENARIA and SUPRIA CT systems, performs equivalently to a legally marketed predicate device (Philips Multislice CT System with Low Dose CT Lung Cancer Screening - K153444).
Here's an analysis based on the provided information:
1. A table of acceptance criteria and the reported device performance
The document does not explicitly state "acceptance criteria" with numerical thresholds set for a new device. Instead, it defines "Image Quality Metric CTQs" (Critical to Quality) as important parameters for lung cancer screening and subsequently compares the subject devices (Hitachi SCENARIA and SUPRIA CT systems with the LCS option) to the predicate device (Philips Brilliance CT 64-channel) based on these metrics. The goal was to prove "substantial equivalence," meaning the new device performs similarly to the predicate.
Here's a table summarizing the image quality metrics and the reported comparative performance:
| Image Quality Metric CTQs | Reason for Inclusion | Reported Device Performance (Comparative) |
|---|---|---|
| CT number accuracy | In a low signal situation such as with low dose LCS, the CT number measured in a nodule may be compromised. In LCS, the CT number may be a reference against potentially calcified nodules. | Demonstrated that CT numbers for all scanners (Hitachi SCENARIA, SUPRIA, and Philips comparison unit) match each other to within ~3 Hounsfield numbers. |
| CT number uniformity | In a low signal situation such as with low dose LCS, maintaining sufficient CT number uniformity throughout the lung and various structures is important for more robust detectability of the nodules. Uniformity is needed to maintain CT number separation between structures. | (Implicitly covered by CT number accuracy and CNR - the comparison asserts overall similar performance without a specific separate uniformity quantification in the summary). The study noted it measured uniformity (variation of CNR and/or mean CT numbers over a range of slices). |
| Image noise (standard deviation) | As dose is reduced, background noise in the image increases. If this noise becomes too large, nodule detectability and sizing measurement may be compromised. | Demonstrated that the variation (standard deviation) of CNR for the phantom test objects is in the range of 6%-8% among the two Hitachi scanners and also for the Philips comparison unit. This suggests comparable noise related to contrast. (Note: standard deviation of CNR is related to image noise). |
| Visual Resolution/Image Artifact | This relates to the evaluation of images to assess their visual resolution using high contrast bar patterns and evaluation of the degree of artifacts (e.g., low signal streaks, beam hardening). These tests are relevant because of the high contrast detection task of relatively small objects for this application. Streak or beam hardening artifacts may obscure pathology and affect CT number accuracy. | Demonstrated that the visibility of small high contrast objects (simulated blood vessels in this phantom) is comparable for all filter/recon combinations among the two Hitachi scanners and for the Philips comparison unit. Beam Hardening Correction is utilized in reconstructions. |
| Contrast to Noise (CNR) | Sufficient Contrast-to-Noise is needed to detect solid and non-solid nodules in the lung. This metric is similar to SNR but accounts for the contrast between an object and the background. GE believes this is the primary figure of merit to evaluate nodule detectability. | Demonstrated that CNR is linearly related among the two Hitachi scanners and also with the Philips comparison unit. The variation (standard deviation) of CNR was 6%-8%. |
Conclusion on Acceptance: Hitachi concluded that the comparison demonstrated "substantial equivalence" based on these metrics, meaning the subject device performs as effectively and safely as the predicate.
2. Sample size used for the test set and the data provenance (e.g., country of origin of the data, retrospective or prospective)
- Sample size: The non-clinical testing was performed using phantom studies. The document specifies repetition and slice counts for the phantom measurements:
- SUPRIA and SCENARIA scanners: 15 repetitions one slice, 15 slices one study.
- Philips scanner: 17 repetitions one slice, 25 slices one study.
- Data provenance: This was a non-clinical bench study comparing CT scanners in a lab setting, not human data. Therefore, country of origin or retrospective/prospective classification (as typically applied to clinical trials) is not applicable. The study was conducted by Hitachi Healthcare Americas.
3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts (e.g., radiologist with 10 years of experience)
- For the non-clinical phantom study, there were no human experts establishing ground truth in the traditional sense of clinical interpretation. The "ground truth" was the physical properties of the phantom and the objective numerical measurements derived from the CT images of the phantom.
- The analysis was done using MATLAB, implying objective quantitative assessment.
4. Adjudication method (e.g., 2+1, 3+1, none) for the test set
- Since the test set involved objective phantom measurements and not human interpretation of clinical images, an adjudication method for expert consensus is not applicable. The measurements were quantitative and compared directly.
5. If a multi reader multi case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance
- No MRMC study was done. The submission states: "Hitachi has determined that additional clinical data for our LCS feature is not needed and that comparative phantom analysis is sufficient to demonstrate substantial equivalence."
- This device is not an AI algorithm adding assistance to human readers. It's a set of low-dose protocols for CT systems that are already cleared. The comparison is between the performance of the CT system with these protocols to a predicate CT system with similar protocols, using phantom measurements. Therefore, the question about AI assistance and reader improvement is not relevant to this submission.
6. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done
- This device is not an algorithm in the sense of a standalone AI diagnostic tool. It is a set of acquisition protocols for a CT scanner. The "standalone" performance in this context would refer to the image quality produced by the CT system using these protocols without human intervention in the acquisition process, which was assessed via the phantom study.
7. The type of ground truth used (expert consensus, pathology, outcomes data, etc.)
- The ground truth for the non-clinical testing was the physical properties of the phantom itself. The phantom contains materials of known density and structures of known size. The measurements (e.g., CT numbers, contrast, small object visibility) derived from the CT images are compared against these known physical properties and also against the performance of the predicate device.
8. The sample size for the training set
- This submission focuses on protocols for existing CT systems, not a new algorithm that requires a "training set" in the context of machine learning or AI. The protocols were developed to take advantage of existing CT system capabilities, and their effectiveness was demonstrated by comparing their image quality metrics to a predicate device. Therefore, a "training set" as commonly understood in AI/ML is not applicable.
9. How the ground truth for the training set was established
- As a training set is not applicable, establishing corresponding ground truth is also not applicable. The protocols themselves were designed based on engineering principles and NEMA XR25 guidelines to optimize dose reduction while maintaining image quality. Their performance was then validated through the comparative phantom study.
§ 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.