(190 days)
Dynamic Visualization II Image Processing is an optional software for the FDX Console, intended to provide optimized image quality over a wide range of patient thicknesses, especially for bariatric imaging. The device is not intended for mammography use.
Fujifilm's FDR D-EVO Flat Panel Detector System (DR-ID600) is a portable digital detector system that interfaces with, and acquires and digitizes x-ray exposures, from standard radiographic svstems. The FDR D-EVO is designed to be used in any environment that would typically use a radiographic cassette for examinations of adults, pediatrics and neonates. The detector models support both wireless and wired/tethered data communication between the detector and the system. Detectors can be placed in a wall bucky for upright exams, a table bucky for recumbent exams, or removed from the bucky for non-grid exams. The Acquisition Workstation for the DR-ID600 is the FDX Console. Dynamic Visualization II Image Processing (DVII) is optional software included in v9.0 of the FDX Console. The FDX Console and Dynamic Visualization II software may be used with Fujifilm DR and CR X-ray systems.
Dynamic Visualization II is an enhanced version of Fujifilm's Dynamic Visualization™ (DV) image processing software. The enhancements made to DVII are designed to provide optimized image quality over a wide range of patient thicknesses, and can be particularly useful when imaging bariatric patients.
DVII uses the same image processing sequence as the predicate DV, but the EDR (Exposure Data Recognition) and MFP (Multi-objective Frequency Processing) algorithms have been modified. When compared to the current EDR and MFP algorithms, the corresponding new algorithms (EDR2, MFP2) have been modified as follows:
EDR2 - to identify and optimize various anatomic structures in an acquired image prior to the subsequent application of contrast and sharpness image processing steps, EDR2 uses a feature recognition method as opposed to conventional EDR's histogram analysis method.
MFP2 - similar to MFP, MFP2 sharpens and balances contrast in anatomic structures in an image after being subject to Exposure Data Recognition. MFP2 uses additional low frequency tables and a combination of automatic and preset dynamic range control operations.
Here's an analysis of the provided information regarding the acceptance criteria and study for the Dynamic Visualization II Image Processing Option:
Acceptance Criteria and Device Performance
The document does not explicitly present a table of acceptance criteria with numerical targets. Instead, the "acceptance criteria" are implied by the comparative image quality evaluation, which aimed to show that the new processing option (DV2) optimizes image quality relative to the predicate (DV), especially for bariatric patients.
Implied Acceptance Criteria and Reported Device Performance:
| Acceptance Criteria (Implied) | Reported Device Performance |
|---|---|
| Image quality optimization for all patient thicknesses | The evaluation "demonstrates that DV2 optimizes image quality for all patients." |
| Improved image quality for bariatric patients | The evaluation "demonstrates that DV2 optimizes image quality for all patients, even when images of larger patients are processed using the proposed, modified device." (This implicitly covers bariatric patients as "larger patients" and notes their specific inclusion in the study design). |
| Device is as safe and effective as the predicate | "FMSU concludes the Dynamic Visualization II Image Processing Option... is as safe and effective as the legally marketed device K153464 and does not raise different questions of safety and effectiveness than K153464." |
| Conformity to voluntary standards (e.g., AAMI/ANSI ES60601-1) | "The conformity to the voluntary standards such as AAMI/ANSI ES60601-1, IEC 60601-1, IEC 60601-1-2, IEC 62304, IEC 62366, IEC 62494-1 and DICOM remains unaffected." |
| Satisfactory verification and validation of improvements | "all verification and validation activities related to the improvements made to the Dynamic Visualization II Image Processing Option were performed and the results were satisfactory." |
Study Details:
-
Sample size used for the test set and the data provenance:
- Sample Size: The document states that "Raw clinical images were processed" and "Randomized image pairs (DV vs. DV2) were then evaluated." However, the exact number of images in the test set is not specified.
- Data Provenance: The data consisted of "Raw clinical images," but the country of origin is not specified. It is stated that "Slightly more than half of the images were of bariatric patients," indicating a focus on a specific patient demographic relevant to the device's indications for use. The study is retrospective as it used "Raw clinical images" that were then processed by both the predicate and proposed algorithms.
-
Number of experts used to establish the ground truth for the test set and the qualifications of those experts:
- Number of Experts: Three readers evaluated the randomized image pairs.
- Qualifications of Experts: The qualifications of these readers are not specified. It is only mentioned that they were "readers," implying they are likely medical professionals, but their specific specialty (e.g., radiologist) and experience level are not provided.
-
Adjudication method for the test set:
- The document states that "Randomized image pairs (DV vs. DV2) were then evaluated by three readers." It does not specify an explicit adjudication method like 2+1 or 3+1. The phrasing suggests individual evaluations by each reader rather than a consensus-building process with formal adjudication rules.
-
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:
- A comparative effectiveness study was performed, focusing on the image processing algorithms (DV vs. DV2) rather than direct AI assistance to human readers. The study compared images processed by the predicate algorithm to images processed by the new algorithm. It was a "comparative image quality evaluation" where "Randomized image pairs (DV vs. DV2) were then evaluated by three readers."
- This was not an MRMC study comparing human readers with vs. without AI assistance. It was a comparison of two different image processing algorithms applied to images, and human readers then evaluated the quality differences between the outputs of these algorithms. Therefore, an effect size of human readers improving with/without AI assistance is not applicable in this context. The study aimed to show the new algorithm itself produces better images.
-
If a standalone (i.e., algorithm only without human-in-the-loop performance) was done:
- The study implicitly involves a standalone performance aspect if you consider the algorithms' ability to process images. The output of the algorithms (processed images) was then directly evaluated by human readers. However, the direct performance of the algorithm without human evaluation is not specifically detailed or presented (e.g., objective metrics of image quality without human perception). The evaluation method involved human readers assessing the output.
-
The type of ground truth used (expert consensus, pathology, outcomes data, etc.):
- The ground truth for the "image quality optimization" was based on the evaluation of three readers (experts). There's no mention of pathology, outcomes data, or a formal expert consensus process to establish ground truth in the sense of a definitive diagnostic label. The evaluation was relative: which processed image (DV or DV2) looked "better" or "optimized."
-
The sample size for the training set:
- The document does not specify the sample size for any training set. The device is described as an enhanced image processing software (Dynamic Visualization II) with modified algorithms (EDR2, MFP2) from a predicate (Dynamic Visualization), not necessarily a machine learning model that requires a distinct training set. While these algorithms were "modified," the process of their development or any explicit "training" of a machine learning model with a defined training set size is not detailed.
-
How the ground truth for the training set was established:
- As no training set is explicitly mentioned or detailed, the method for establishing its ground truth is not provided. The changes are described as modifications to existing image processing algorithms (EDR and MFP), suggesting an engineering or algorithmic refinement process rather than, for example, supervised machine learning requiring labeled training data.
§ 892.1680 Stationary x-ray system.
(a)
Identification. A stationary x-ray system is a permanently installed diagnostic system intended to generate and control x-rays for examination of various anatomical regions. This generic type of device may include signal analysis and display equipment, patient and equipment supports, component parts, and accessories.(b)
Classification. Class II (special controls). A radiographic contrast tray or radiology diagnostic kit intended for use with a stationary x-ray system only is exempt from the premarket notification procedures in subpart E of part 807 of this chapter subject to the limitations in § 892.9.