Search Results

The uMR Jupiter system is indicated for use as a magnetic resonance diagnostic device (MRDD) that produces sagittal, transverse, coronal, and oblique cross sectional images, and spectroscopic images, and that display internal anatomical structure and/or function of the head, body and extremities.

These images and the physical parameters derived from the images when interpreted by a trained physician yield information that may assist the diagnosis. Contrast agents may be used depending on the region of interest of the scan.

The device is intended for patients > 20 kg/44 lbs.

uMR Jupiter is a 5T superconducting magnetic resonance diagnostic device with a 60cm size patient bore and 8 channel RF transmit system. It consists of components such as magnet, RF power amplifier, RF coils, gradient power amplifier, gradient coils, patient table, spectrometer, computer, equipment cabinets, power distribution system, internal communication system, and vital signal module etc. uMR Jupiter is designed to conform to NEMA and DICOM standards.

The modification performed on the uMR Jupiter in this submission is due to the following changes that include:

1. Addition of RF coils: SuperFlex Large - 24 and Foot & Ankle Coil - 24.

2. Addition of applied body part for certain coil: SuperFlex Small-24 (add imaging of ankle).

3. Addition and modification of pulse sequences:

   • a) New sequences: fse_wfi, gre_fsp_c (3D), gre_bssfp_ucs, epi_fid(3D), epi_dti_msh.
   • b) Added Associated options for certain sequences: asl_3d (add mPLD) (Only output original images and no quantification images are output), gre_fsp_c (add Cardiac Cine, Cardiac Perfusion, PSIR, Cardiac mapping), gre_quick(add WFI, MRCA), gre_bssfp(add Cardiac Cine, Cardiac mapping), epi_dwi(add IVIM) (Only output original images and no quantification images are output).

4. Addition of function: EasyScan, EasyCrop, t-ACS, QScan, tFAST, DeepRecon and WFI.

5. Addition of workflow: EasyFACT.

This FDA 510(k) summary (K250246) for the uMR Jupiter provides details on several new AI-assisted features. Here's a breakdown of the acceptance criteria and the study that proves the device meets them, based on the provided text:

Important Note: The document is not a MRMC study comparing human readers with and without AI. Instead, it focuses on the performance of individual AI modules and their integration into the MRI system, often verified by radiologists' review of image quality.

Acceptance Criteria and Reported Device Performance

The document presents acceptance criteria implicitly through the "Test Result" or "Performance Verification" sections for each AI feature. The "Performance" column below summarizes the device's reported achievement for these criteria.

1. "A better consistency between t-ACS and the reference than that between CS and the reference was shown in all t-ACS application types."
2. "No large structural difference appeared between t-ACS and the reference in all t-ACS application types."
3. "The motion-time curves and Bland-Altman analysis showed the consistency between t-ACS and the reference based on simulated and real acquired data in all t-ACS application types."
   Overall: "The t-ACS on uMR Jupiter was shown to perform better than traditional Compressed Sensing in the sense of discrepancy from fully sampled images and PSNR using images from various age groups, BMIs, ethnicities and pathological variations. The structure measurements on paired images verified that same structures of t-ACS and reference were significantly the same. And t-ACS integration tests in two applications proved that t-ACS had good agreement with the reference." |
   | DeepRecon | Expected to provide image de-noising and super-resolution, resulting in diagnostic quality images, with equivalent or higher scores than reference images in terms of diagnostic quality. | "The DeepRecon has been validated to provide image de-nosing and super-resolution processing using various ethnicities, age groups, BMIs, and pathological variations. In addition, DeepRecon images were evaluated by American Board of Radiologists certificated physicians, covering a range of protocols and body parts. The evaluation reports from radiologists verified that DeepRecon meets the requirements of clinical diagnosis. All DeepRecon images were rated with equivalent or higher scores in terms of diagnosis quality." |
   | EasyFACT | Expected to effectively automate ROI placement and numerical statistics for FF and R2* values, with results subjectively evaluated as effective. | "The subjective evaluation method was used [to verify effectiveness]." "The proposal of algorithm acceptance criteria and score processing are conducted by the licensed physicians with U.S. credentials." (Implied successful verification from context) |
   | EasyScan | Pass criteria of 99.3% for automatic slice group positioning, meeting safety and effectiveness requirements. | "The pass criteria of EasyScan feature is 99.3%, and the results evaluated by the licenced MRI technologist with U.S. credentials. Therefore, EasyScan meets the criteria for safety and effectiveness, and EasyScan can meet the requirements for automatic positioning locates slice groups." (Implied reaching or exceeding 99.3%.) |
   | EasyCrop | Pass criteria of 100% for automatic image cropping, meeting safety and effectiveness requirements. | "The pass criteria of EasyCrop feature is 100%, and the results evaluated by the licenced MRI technologist with U.S. credentials. Therefore, EasCrop meets the criteria for safety and effectiveness, and EasCrop can meet the requirements for automatic cropping." (Implied reaching or exceeding 100%.) |

Study Details

1. Sample sizes used for the test set and the data provenance:

   • WFI: 144 cases from 28 volunteers. Data collected from UIH Jupiter. "Completely separated from the previous mentioned training dataset by collecting from different volunteers and during different time periods." (Retrospective for testing, though specific country of origin beyond "UIH MRI systems" is not explicitly stated for testing data, training data has "Asian" majority.)
   • t-ACS: 35 subjects (data from 76 volunteers used for overall training/validation/test split). Test data collected independently from the training data, with separated subjects and during different time periods. "White," "Black," and "Asian" ethnicities mentioned, implying potentially multi-country or diverse internal dataset.
   • DeepRecon: 20 subjects (2216 cases). "Diverse demographic distributions" including "White" and "Asian" ethnicities. "Collecting testing data from various clinical sites and during separated time periods."
   • EasyFACT: 5 subjects. "Data were acquired from 5T magnetic resonance imaging equipment from UIH," and "Asia" ethnicity is listed.
   • EasyScan: 30 cases from 18 "Asia" subjects (initial testing); 40 cases from 8 "Asia" subjects (validation on uMR Jupiter system).
   • EasyCrop: 5 subjects. "Data were acquired from 5T magnetic resonance imaging equipment from UIH," and "Asia" ethnicity is listed.

   Data provenance isn't definitively "retrospective" or "prospective" for the test sets, but the emphasis on "completely separated" and "independent" from training data collected at "different time periods" suggests these were distinct, potentially newly acquired or curated sets for evaluation. The presence of multiple ethnicities (White, Black, Asian) suggests potentially broader geographical origins than just China where the company is based, or a focus on creating diverse internal datasets.

2. Number of experts used to establish the ground truth for the test set and the qualifications of those experts:

   • WFI: Three U.S. certificated radiologists. (Qualifications: U.S. board-certified radiologists).
   • t-ACS: No separate experts establishing ground truth for the test set performance evaluation are mentioned beyond the quantitative metrics (MAE, PSNR, SSIM) compared against "fully sampled images" (reference/ground truth). The document states that fully-sampled k-space data transformed into image domain served as the reference.
   • DeepRecon: American Board of Radiologists certificated physicians. (Qualifications: American Board of Radiologists certificated physicians).
   • EasyFACT: Licensed physicians with U.S. credentials. (Qualifications: Licensed physicians with U.S. credentials).
   • EasyScan: Licensed MRI technologist with U.S. credentials. (Qualifications: Licensed MRI technologist with U.S. credentials).
   • EasyCrop: Licensed MRI technologist with U.S. credentials. (Qualifications: Licensed MRI technologist with U.S. credentials).

3. Adjudication method (e.g. 2+1, 3+1, none) for the test set:

   The document does not explicitly state an adjudication method (like 2+1 or 3+1) for conflict resolution among readers. For WFI, DeepRecon, EasyFACT, EasyScan, and EasyCrop, it implies a consensus or majority opinion model based on the "evaluation reports from radiologists/technologists." For t-ACS, the evaluation of the algorithm's output is based on quantitative metrics against a reference image ground truth.

4. 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, a traditional MRMC comparative effectiveness study where human readers interpret cases with AI assistance versus without AI assistance was not described. The studies primarily validated the AI features' standalone performance (e.g., image quality, accuracy of automated functions) or their output's equivalence/superiority to traditional methods, often through expert review of the AI-generated images. Therefore, no effect size of human reader improvement with AI assistance is provided.

5. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done:

   Yes, standalone performance was the primary focus for most AI features mentioned, though the output was often subject to human expert review.

   • WFI: The AI network provides initialization for the RIPE algorithm. The output image quality was then reviewed by radiologists.
   • t-ACS: Performance was evaluated quantitatively against fully sampled images (reference/ground truth), indicating a standalone algorithm evaluation.
   • DeepRecon: Evaluated based on images processed by the algorithm, with expert review of the output images.
   • EasyFACT, EasyScan, EasyCrop: These are features that automate parts of the workflow. Their output (e.g., ROI placement, slice positioning, cropping) was evaluated, often subjectively by experts, but the automation itself is algorithm-driven.

6. The type of ground truth used (expert consensus, pathology, outcomes data, etc.):

   • WFI: Expert consensus/review by three U.S. certificated radiologists for "clinical diagnosis" quality. No "ground truth" for water-fat separation accuracy itself is explicitly stated, but the problem being solved (water-fat swap artifacts) implies the improved stability of the algorithm's output.
   • t-ACS: "Fully-sampled k-space data were collected and transformed into image domain as reference." This serves as the "true" or ideal image for comparison, not derived from expert interpretation or pathology.
   • DeepRecon: "Multiple-averaged images with high-resolution and high SNR were collected as the ground-truth images." Expert review confirms diagnostic quality of processed images.
   • EasyFACT: Subjective evaluation by licensed physicians with U.S. credentials, implying their judgment regarding the correctness of ROI placement and numerical statistics.
   • EasyScan: Evaluation by a licensed MRI technologist with U.S. credentials against the "correctness" of automatic slice positioning.
   • EasyCrop: Evaluation by a licensed MRI technologist with U.S. credentials against the "correctness" of automatic cropping.

7. The sample size for the training set:

   • WFI AI module: 59 volunteers (2604 cases). Each scanned for multiple body parts and WFI protocols.
   • t-ACS AI module: Not specified as a distinct number, but "collected from a variety of anatomies, image contrasts, and acceleration factors... resulting in a large number of cases." The overall dataset for training, validation, and testing was 76 volunteers.
   • DeepRecon: 317 volunteers.
   • EasyFACT, EasyScan, EasyCrop: "The training data used for the training of the EasyFACT algorithm is independent of the data used to test the algorithm." For EasyScan and EasyCrop, it states "The testing dataset was collected independently from the training dataset," but does not provide specific training set sizes for these workflow features.

8. How the ground truth for the training set was established:

   • WFI AI module: The AI network was trained to provide accurate initialization for the RIPE algorithm. The document implies that the RIPE algorithm itself with human oversight or internal validation would have been used to establish correct water/fat separation for training.
   • t-ACS AI module: "Fully-sampled k-space data were collected and transformed into image domain as reference." This served as the ground truth against which the AI was trained to reconstruct undersampled data.
   • DeepRecon: "The multiple-averaged images with high-resolution and high SNR were collected as the ground-truth images." This indicates that high-quality, non-denoised, non-super-resolved images were used as the ideal target for the AI.
   • EasyFACT, EasyScan, EasyCrop: Not explicitly detailed beyond stating that training data ground truth was established to enable the algorithms for automatic ROI placement, slice group positioning, and image cropping, respectively. It implies a process of manually annotating or identifying the correct ROIs/positions/crops on training data for the AI to learn from.

Ask a specific question about this device

The uMR Ultra system is indicated for use as a magnetic resonance diagnostic device (MRDD) that produces sagittal, transverse, coronal, and oblique cross sectional images, and spectroscopic images, and that display internal anatomical structure and/or function of the head, body and extremities. These images and the physical parameters derived from the images when interpreted by a trained physician yield information that may assist the diagnosis. Contrast agents may be used depending on the region of interest of the scan.

uMR Ultra is a 3T superconducting magnetic resonance diagnostic device with a 70cm size patient bore and 2 channel RF transmit system. It consists of components such as magnet, RF power amplifier, RF coils, gradient power amplifier, gradient coils, patient table, spectrometer, computer, equipment cabinets, power distribution system, internal communication system, and vital signal module etc. uMR Ultra is designed to conform to NEMA and DICOM standards.

Here's a breakdown of the acceptance criteria and study details for the uMR Ultra device, based on the provided FDA 510(k) clearance letter.

1. Table of Acceptance Criteria and Reported Device Performance

Given the nature of the document, which focuses on device clearance, multiple features are discussed. I will present the acceptance criteria and results for the AI-powered features, as these are the most relevant to the "AI performance" aspect.

Acceptance Criteria and Device Performance for AI-Enabled Features

• ACS has higher SNR than CS.
• ACS has higher (standard deviation (SD) / mean value(S)) values than CS.
• Bland-Altman analysis of image intensities acquired using fully sampled and ACS shown with less than 1% bias and all sample points falls in the 95% confidence interval.
• Measurement differences on ACS and fully sampled images of same structures under 5% is acceptable.
• Radiologists rate all ACS images with equivalent or higher scores in terms of diagnosis quality. | - Pass
• Pass
• Pass
• Pass
• Pass
• Verified that ACS meets the requirements of clinical diagnosis. All ACS images were rated with equivalent or higher scores in terms of diagnosis quality. |
  | DeepRecon | - DeepRecon images achieve higher SNR compared to NADR images.
• Uniformity difference between DeepRecon images and NADR images under 5%.
• Intensity difference between DeepRecon images and NADR images under 5%.
• Measurements on NADR and DeepRecon images of same structures, measurement difference under 5%.
• Radiologists rate all DeepRecon images with equivalent or higher scores in terms of diagnosis quality. | - NADR: 343.63, DeepRecon: 496.15 (PASS)
• 0.07% (PASS)
• 0.2% (PASS)
• 0% (PASS)
• Verified that DeepRecon meets the requirements of clinical diagnosis. All DeepRecon images were rated with equivalent or higher scores in terms of diagnosis quality. |
  | EasyScan | No Fail cases and auto position success rate P1/(P1+P2+F) exceeds 80%.
  (P1: Pass with auto positioning; P2: Pass with user adjustment; F: Fail) | 99.6% |
  | t-ACS | - AI prediction (AI module output) much closer to reference compared to AI module input images.
• Better consistency between t-ACS and reference than between CS and reference.
• No large structural difference appeared between t-ACS and reference.
• Motion-time curves and Bland-Altman analysis consistency between t-ACS and reference. | - Pass
• Pass
• Pass
• Pass |
  | AiCo | - AiCo images exhibit improved PSNR and SSIM compared to the originals.
• No significant structural differences from the gold standard.
• Radiologists confirm image quality is diagnostically acceptable, fewer motion artifacts, and greater benefits for clinical diagnosis. | - Pass
• Pass
• Confirmed. |
  | SparkCo | - Average detection accuracy needs to be > 90%.
• Average PSNR of spark-corrected images needs to be higher than spark images.
• Spark artifacts need to be reduced or corrected after enabling SparkCo. | - 94%
• 1.6 higher
• Successfully corrected |
  | ImageGuard | Success rate P/(P+F) exceeds 90%.
  (P: Pass if prompt appears for motion / no prompt for no motion; F: Fail if prompt doesn't appear for motion / prompt appears for no motion) | 100% |
  | EasyCrop | No Fail cases and pass rate P1/(P1+P2+F) exceeds 90%.
  (P1: Other peripheral tissues cropped, meets user requirements; P2: Cropped images don't meet user requirements, but can be re-cropped; F: EasyCrop fails or original images not saved) | 100% |
  | EasyFACT | Satisfied and Acceptable ratio (S+A)/(S+A+F) exceeds 95%.
  (S: All ROIs placed correctly; A: Fewer than five ROIs placed correctly; F: ROIs positioned incorrectly or none placed) | 100% |
  | Auto TI Scout | Average frame difference between auto-calculated TI and gold standard is ≤ 1 frame, and maximum frame difference is ≤ 2 frames. | Average: 0.37-0.44 frames, Maximum: 1-2 frames (PASS) |
  | Inline MOCO | Average Dice coefficient of the left ventricular myocardium after motion correction is > 0.87. | Cardiac Perfusion Images: 0.92
  Cardiac Dark Blood Images: 0.96 |
  | Inline ED/ES Phases Recognition | The average error between the phase indices calculated by the algorithm for the ED and ES of test data and the gold standard phase indices does not exceed 1 frame. | 0.13 frames |
  | Inline ECV | No failure cases, satisfaction rate S/(S+A+F) > 95%.
  (S: Segmentation adheres to myocardial boundary, blood pool ROI correct; A: Small missing/redundant areas but blood pool ROI correct; F: Myocardial mask fails or blood pool ROI incorrect) | 100% |
  | EasyRegister (Height Estimation) | PH5 (Percentage of height error within 5%); PH15 (Percentage of height error within 15%); MEAN_H (Average error of height). (Specific numerical criteria not explicitly stated beyond these metrics) | PH5: 92.4%
  PH15: 100%
  MEAN_H: 31.53mm |
  | EasyRegister (Weight Estimation) | PW10 (Percentage of weight error within 10%); PW20 (Percentage of weight error within 20%); MEAN_W (Average error of weight). (Specific numerical criteria not explicitly stated beyond these metrics) | PW10: 68.64%
  PW20: 90.68%
  MEAN_W: 6.18kg |
  | EasyBolus | No Fail cases and success rate P1+P2/(P1+P2+F) exceeds 100%.
  (P1: Monitoring point positioning meets user requirements, frame difference ≤ 1 frame; P2: Monitoring point positioning meets user requirements, frame difference = 2 frames; F: Auto position fails or frame difference > 2 frames) | P1: 80%
  P2: 20%
  Total Failure Rate: 0%
  Pass: 100% |

For the rest of the questions, I will consolidate the information where possible, as some aspects apply across multiple AI features.

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

• ACS: 749 samples from 25 volunteers. Diverse demographic distributions covering various genders, age groups, ethnicity (White, Asian, Black), and BMI (Underweight, Healthy, Overweight/Obesity). Data collected from various clinical sites during separated time periods.
• DeepRecon: 25 volunteers (nearly 2200 samples). Diverse demographic distributions covering various genders, age groups, ethnicity (White, Asian, Black), and BMI. Data collected from various clinical sites during separated time periods.
• EasyScan: 444 cases from 116 subjects. Diverse demographic distributions covering various genders, age groups, and ethnicities. Data acquired from UIH MRI equipment (1.5T and 3T). Data provenance not explicitly stated (e.g., country of origin), but given the company location (China) and "U.S. credentials" for evaluators, it likely includes data from both. The document states "The testing dataset was collected independently from the training dataset".
• t-ACS: 1173 cases from 60 volunteers. Diverse demographic distributions covering various genders, age groups, ethnicities (White, Black, Asian) and BMI. Data acquired by uMR Ultra scanners. Data provenance not explicitly stated, but implies global standard testing.
• AiCo: 218 samples from 24 healthy volunteers. Diverse demographic distributions covering various genders, age groups, BMI (Under/healthy weight, Overweight/Obesity), and ethnicity (White, Black, Asian). Data provenance not explicitly stated.
• SparkCo: 59 cases from 15 patients for real-world spark raw data testing. Diverse demographic distributions including gender, age, BMI (Underweight, Healthy, Overweight, Obesity), and ethnicity (Asian, "N.A." for White, implying not tested as irrelevant). Data acquired by uMR 1.5T and uMR 3T scanners.
• ImageGuard: 191 cases from 80 subjects. Diverse demographic distributions covering various genders, age groups, and ethnicities (White, Black, Asian). Data acquired from UIH MRI equipment (1.5T and 3T).
• EasyCrop: Not explicitly stated as "subjects" vs. "cases," but tested on 5 intended imaging body parts. Sample size (N=65) implies 65 cases/scans, potentially from 65 distinct subjects or fewer if subjects had multiple scans. Diverse demographic distributions covering various genders, age groups, ethnicity (Asian, Black, White). Data acquired from UIH MRI equipment (1.5T and 3T).
• EasyFACT: 25 cases from 25 volunteers. Diverse demographic distributions covering various genders, age groups, weight, and ethnicity (Asian, White, Black).
• Auto TI Scout: 27 patients. Diverse demographic distributions covering various genders, age groups, ethnicity (Asian, White), and BMI. Data acquired from 1.5T and 3T scanners.
• Inline MOCO: Cardiac Perfusion Images: 105 cases from 60 patients. Cardiac Dark Blood Images: 182 cases from 33 patients. Diverse demographic distributions covering age, gender, ethnicity (Asian, White, Black, Hispanic), BMI, field strength, and disease conditions (Positive, Negative, Unknown).
• Inline ED/ES Phases Recognition: 95 cases from 56 volunteers, covering various genders, age groups, field strength, disease conditions (NOR, MINF, DCM, HCM, ARV), and ethnicity (Asian, White, Black).
• Inline ECV: 90 images from 28 patients. Diverse demographic distributions covering gender, age, BMI, field strength, ethnicity (Asian, White), and health status (Negative, Positive, Unknown).
• EasyRegister (Height/Weight Estimation): 118 cases from 63 patients. Diverse ethnic groups (Chinese, US, France, Germany).
• EasyBolus: 20 subjects. Diverse demographic distributions covering gender, age, field strength, and ethnicity (Asia).

Data Provenance (Retrospective/Prospective and Country of Origin):
The document states "The testing dataset was collected independently from the training dataset, with separated subjects and during different time periods." This implies a prospective collection for validation that is distinct from the training data. For ACS and DeepRecon, it explicitly mentions "US subjects" for some evaluations, but for many features, the specific country of origin for the test set is not explicitly stated beyond "diverse ethnic groups" or "Asian" which could be China (where the company is based) or other Asian populations. The use of "U.S. board-certified radiologists" and "licensed MRI technologist with U.S. credentials" for evaluation suggests the data is intended to be representative of, or directly includes, data relevant to the U.S. clinical context.

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

• ACS & DeepRecon: Evaluated by "American Board of Radiologists certificated physicians" (plural, implying multiple, at least 2). Not specified how many exactly, but strong qualifications.
• EasyScan, ImageGuard, EasyCrop, EasyBolus: Evaluated by "licensed MRI technologist with U.S. credentials." For EasyBolus, it specifies "certified professionals in the United States." Number not explicitly stated beyond "the" technologist/professionals, but implying multiple for robust evaluation.
• Inline MOCO & Inline ECV: Ground truth annotations done by a "well-trained annotator" and "finally, all ground truth was evaluated by three licensed physicians with U.S. credentials." This indicates a 3-expert consensus/adjudication.
• SparkCo: "One experienced evaluator" for subjective image quality improvement.
• For other features (t-ACS, EasyFACT, Auto TI Scout, Inline ED/ES Phases Recognition, EasyRegister), the ground truth seems to be based on physical measurements (for EasyRegister) or computational metrics (for t-ACS based on fully-sampled images, and for accuracy of ROI placement against defined standards), rather than human expert adjudication for ground truth.

4. Adjudication Method (e.g., 2+1, 3+1, none) for the Test Set

• Inline MOCO & Inline ECV: "Evaluated by three licensed physicians with U.S. credentials." This implies a 3-expert consensus method for ground truth establishment.
• ACS, DeepRecon, AiCo: "Evaluated by American Board of Radiologists certificated physicians" (plural). While not explicitly stated as 2+1 or 3+1, it suggests a multi-reader review, where consensus was likely reached for the reported diagnostic quality.
• SparkCo: "One experienced evaluator" was used for subjective evaluation, implying no formal multi-reader adjudication for this specific metric.
• For features like EasyScan, ImageGuard, EasyCrop, EasyBolus (evaluated by MRI technologists) and those relying on quantitative metrics against a reference (t-ACS, EasyFACT, Auto TI Scout, EasyRegister, Inline ED/ES Phases Recognition), the "ground truth" is either defined by the system's intended function (e.g., correct auto-positioning) or a mathematically derived reference, so a traditional human adjudication method is not applicable in the same way as for diagnostic image interpretation.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was Done

The document does not explicitly state that a formal MRMC comparative effectiveness study was performed to quantify the effect size of how much human readers improve with AI vs. without AI assistance.

Instead, the evaluations for ACS, DeepRecon, and AiCo involve "American Board of Radiologists certificated physicians" who "verified that [AI feature] meets the requirements of clinical diagnosis. All [AI feature] images were rated with equivalent or higher scores in terms of diagnosis quality." For AiCo, they confirmed images "exhibit fewer motion artifacts and offer greater benefits for clinical diagnosis." This is a qualitative assessment of diagnostic quality by experts, but not a comparative effectiveness study in the sense of measuring reader accuracy or confidence change with AI assistance.

6. If a Standalone (i.e., algorithm only without human-in-the-loop performance) was Done

Yes, for many of the AI-enabled features, a standalone performance evaluation was conducted:

• ACS: Performance was evaluated by comparing quantitative metrics (NRMSE, SNR, Resolution, Contrast, Uniformity, Structure Measurement) against fully-sampled images or CS. This is a standalone evaluation.
• DeepRecon: Quantitative metrics (SNR, uniformity, contrast, structure measurement) were compared between DeepRecon and NADR (without DeepRecon) images. This is a standalone evaluation.
• t-ACS: Quantitative tests (MAE, PSNR, SSIM, structural measurements, motion-time curves) were performed comparing t-ACS and CS results against a reference. This is a standalone evaluation.
• AiCo: PSNR and SSIM values were quantitatively compared, and structural dimensions were assessed, between AiCo processed images and original/motionless reference images. This is a standalone evaluation.
• SparkCo: Spark detection accuracy was calculated, and PSNR of spark-corrected images was compared to original spark images. This is a standalone evaluation.
• Inline MOCO: Evaluated using Dice coefficient, a quantitative metric for segmentation accuracy. This is a standalone evaluation.
• Inline ED/ES Phases Recognition: Evaluated by quantifying the error between algorithm output and gold standard phase indices. This is a standalone evaluation.
• Inline ECV: Evaluated by quantitative scoring for segmentation accuracy (S, A, F criteria). This is a standalone evaluation.
• EasyRegister (Height/Weight): Evaluated by quantitative error metrics (PH5, PH15, MEAN_H; PW10, PW20, MEAN_W) against physical measurements. This is a standalone evaluation.

Features like EasyScan, ImageGuard, EasyCrop, and EasyBolus involve automated workflow assistance where the direct "diagnostic" outcome isn't solely from the algorithm, but the automated function's performance is evaluated in a standalone manner against defined success criteria.

7. The Type of Ground Truth Used

The type of ground truth varies depending on the specific AI feature:

• Reference/Fully-Sampled Data:
  • ACS, DeepRecon, t-ACS, AiCo: Fully-sampled k-space data transformed to image space served as "ground-truth" for training and as a reference for quantitative performance metrics in testing. For AiCo, "motionless data" served as gold standard.
  • SparkCo: Simulated spark artifacts generated from "spark-free raw data" provided ground truth for spark point locations in training.
• Expert Consensus/Subjective Evaluation:
  • ACS, DeepRecon, AiCo: "American Board of Radiologists certificated physicians" provided qualitative assessment of diagnostic image quality ("equivalent or higher scores," "diagnostically acceptable," "fewer motion artifacts," "greater benefits for clinical diagnosis").
  • EasyScan, ImageGuard, EasyCrop, EasyBolus: "Licensed MRI technologist with U.S. credentials" or "certified professionals in the United States" performed subjective evaluation against predefined success criteria for the workflow functionality.
  • SparkCo: One experienced evaluator for subjective image quality improvement.
• Anatomical/Physiological Measurements / Defined Standards:
  • EasyFACT: Defined criteria for ROI placement within liver parenchyma, avoiding borders/vascular structures.
  • Auto TI Scout, Inline ED/ES Phases Recognition: Gold standard phase indices were presumably established by expert review or a reference method.
  • Inline MOCO & Inline ECV: Ground truth for cardiac left ventricular myocardium segmentation was established by a "well-trained annotator" and "evaluated by three licensed physicians with U.S. credentials" (expert consensus based on anatomical boundaries).
  • EasyRegister (Height/Weight Estimation): "Precisely measured height/weight value" using "physical examination standards."

8. The Sample Size for the Training Set

• ACS: 1,262,912 samples (collected from variety of anatomies, image contrasts, and acceleration factors, scanned by UIH MRI systems).
• DeepRecon: 165,837 samples (collected from 264 volunteers, scanned by UIH MRI systems for multiple body parts and clinical protocols).
• EasyScan: Training data collection not explicitly detailed in the same way as ACS/DeepRecon (refers to "collected independently from the training dataset").
• t-ACS: Datasets collected from 108 volunteers ("large number of samples").
• AiCo: 140,000 images collected from 114 volunteers across multiple body parts and clinical protocols.
• SparkCo: 24,866 spark slices generated from 61 cases collected from 10 volunteers.
• EasyFACT, Auto TI Scout, Inline MOCO, Inline ED/ES Phases Recognition, Inline ECV, EasyRegister, EasyBolus: The document states that training data was independent of testing data but does not provide specific sample sizes for the training datasets for these features.

9. How the Ground Truth for the Training Set was Established

• ACS, DeepRecon, t-ACS, AiCo: "Fully-sampled k-space data were collected and transformed to image space as the ground-truth." For DeepRecon specifically, "multiple-averaged images with high-resolution and high SNR were collected as the ground-truth images." For AiCo, "motionless data" served as gold standard. All training data were "manually quality controlled."
• SparkCo: "The training dataset... was generated by simulating spark artifacts from spark-free raw data... with the corresponding ground truth (i.e., the location of spark points)."
• Inline MOCO & Inline ECV: The document states "all ground truth was annotated by a well-trained annotator. The annotator used an interactive tool to observe the image, and then labeled the left ventricular myocardium in the image."
• For EasyScan, EasyFACT, Auto TI Scout, Inline ED/ES Phases Recognition, EasyRegister, and EasyBolus training ground truth establishment is not explicitly detailed, only that the testing data was independent of the training data. For EasyRegister, it implies physical measurements were the basis for ground truth.

Ask a specific question about this device

The uMR 680 system is indicated for use as a magnetic resonance diagnostic device (MRDD) that produces sagittal, transverse, coronal, and oblique cross sectional images, and spectroscopic images, and that display internal anatomical structure and/or function of the head, body and extremities.

These images and the physical parameters derived from the images when interpreted by a trained physician yield information that may assist the diagnosis. Contrast agents may be used depending on the region of interest of the scan.

The uMR 680 is a 1.5T superconducting magnetic resonance diagnostic device with a 70cm size patient bore. It consists of components such as magnet, RF power amplifier, RF coils, gradient power amplifier, gradient coils, patient table, spectrometer, computer, equipment cabinets, power distribution system, internal communication system, and vital signal module etc. The uMR 680 Magnetic Resonance Diagnostic Device is designed to conform to NEMA and DICOM standards.

This traditional 510(k) is to request modifications for the cleared uMR 680(K240744). The modifications performed on the uMR 680 in this submission are due to the following changes that include:
(1) Addition of RF coils and corresponding accessories: Breast Coil -12, Biopsy Configuration, Head Coil-16, Positioning Couch-top, Coil Support.
(2) Deletion of VSM (Wireless UIH Gating Unit REF 453564324621, ECG module Ref 989803163121, SpO2 module Ref 989803163111).
(3) Modification of the dimensions of Detachable table: from width 826mm, height 880mm,2578mm to width 810mm, height 880mm, length 2505mm.
(4) Addition and modification of pulse sequences
a) New sequences: gre_snap, gre_quick_4dncemra, gre_pass, gre_mtp, gre_trass, epi_dwi_msh, epi_dti_msh, svs_hise.
b) Added associated options for certain sequences: fse(add Silicone-Only Imaging, MicroView, MTC, MultiBand), fse_arms(add Silicone-Only Imaging), fse_ssh(add Silicone-Only Imaging), fse_mx(add CEST, T1rho, MicroView, MTC), fse_arms_dwi(add MultiBand), asl_3d(add multi-PLD), gre(add T1rho, MTC, output phase image), gre_fsp(add FSP+), gre_bssfp(add CASS, TI Scout), gre_fsp_c(add 3D LGE, DB/GB PSIR), gre_bssfp_ucs(add real time cine), gre_fq(add 4D Flow), epi_dwi(add IVIM), epi_dti(add DKI, DSI).
c) Added additional accessory equipment required for certain sequences: gre_bssfp(add Virtual ECG Trigger).
d) Name change of certain sequences: gre_fine(old name: gre_bssfp_fi).
e) Added applicable body parts: gre_ute, gre_fine, fse_mx.
(5) Addition of imaging reconstruction methods: AI-assisted Compressed Sensing (ACS), Spark artifact Correction (SparkCo).
(6) Addition of imaging processing methods: Inline Cardiac Function, Inline ECV, Inline MRS, Inline MOCO, 4D Flow, SNAP, CEST, T1rho, FSP+, CASS, PASS, MTP.
(7) Addition of workflow features: TI Scout, EasyCrop, ImageGuard, Mocap, EasyFACT, Auto Bolus tracker, Breast Biopsy and uVision.
(8) Modification of workflow features: EasyScan(add applicable body parts)

The modification does not affect the intended use or alter the fundamental scientific technology of the device.

The provided FDA 510(k) clearance letter and summary for the uMR 680 Magnetic Resonance Imaging System outlines performance data for several new features and algorithms.

Here's an analysis of the acceptance criteria and the studies that prove the device meets them for the AI-assisted Compressed Sensing (ACS), SparkCo, Inline ED/ES Phases Recognition, and Inline MOCO algorithms.

1. Table of Acceptance Criteria and Reported Device Performance

• Age: 0.92-0.93
• Gender: 0.92
• Ethnicity: 0.91-0.92
• BMI: 0.91-0.95
• Magnetic field strength: 0.92-0.93
• Disease conditions: 0.91-0.93 |
  | | Dice Coefficient (Left Ventricular Myocardium after Motion Correction) Cardiac Dark Blood Images | The average Dice coefficient of the left ventricular myocardium after motion correction is greater than 0.87. | The average Dice coefficient of the left ventricular myocardium after motion correction is 0.96, which is greater than 0.87. Subgroup analysis also showed good generalization:
• Age: 0.95-0.96
• Gender: 0.96
• Ethnicity: 0.95-0.96
• BMI: 0.96-0.98
• Magnetic field strength: 0.96
• Disease conditions: 0.96-0.97 |

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

• AI-assisted Compressed Sensing (ACS):
  • Sample Size: 1724 samples from 35 volunteers.
  • Data Provenance: Diverse demographic distributions (gender, age groups, ethnicity, BMI) covering various clinical sites and separated time periods. Implied to be prospective or a carefully curated retrospective set, collected specifically for validation on the uMR 680 system, and independent of training data.
• SparkCo:
  • Simulated Spark Testing Dataset: 159 spark slices (generated from spark-free raw data).
  • Real-world Spark Testing Dataset: 59 cases from 15 patients.
  • Data Provenance: Real-world data acquired from uMR 1.5T and uMR 3T scanners, covering representative clinical protocols. The report specifies "Asian" for 100% of the real-world dataset's ethnicity, noting that performance is "irrelevant with human ethnicity" due to the nature of spark signal detection. This is retrospective data.
• Inline ED/ES Phases Recognition:
  • Sample Size: 95 cases from 56 volunteers.
  • Data Provenance: Includes various ages, genders, field strengths (1.5T, 3.0T), disease conditions (NOR, MINF, DCM, HCM, ARV), and ethnicities (Asian, White, Black). The data is independent of the training data. Implied to be retrospective from UIH MRI systems.
• Inline MOCO:
  • Sample Size: 287 cases in total (105 cardiac perfusion images from 60 patients, 182 cardiac dark blood images from 33 patients).
  • Data Provenance: Acquired from 1.5T and 3T magnetic resonance imaging equipment from UIH. Covers various ages, genders, ethnicities (Asian, White, Black, Hispanic), BMI, field strengths (1.5T, 3.0T), and disease conditions (Positive, Negative, Unknown). The data is independent of the training data. Implied to be retrospective from UIH MRI systems.

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

• AI-assisted Compressed Sensing (ACS):
  • Number of Experts: More than one (plural "radiologists" used).
  • Qualifications: American Board of Radiologists certificated physicians.
• SparkCo:
  • Number of Experts: One expert for real-world SparkCo evaluation.
  • Qualifications: "one experienced evaluator." (Specific qualifications like board certification or years of experience are not provided for this specific evaluator).
• Inline ED/ES Phases Recognition:
  • Number of Experts: Not explicitly stated for ground truth establishment ("gold standard phase indices"). It implies a single, established method or perhaps a consensus by a team, but details are missing.
• Inline MOCO:
  • Number of Experts: Three licensed physicians.
  • Qualifications: U.S. credentials.

4. Adjudication Method for the Test Set

• AI-assisted Compressed Sensing (ACS): Not explicitly stated, but implies individual review by "radiologists" to rate diagnostic quality.
• SparkCo: For the real-world dataset, evaluation by "one experienced evaluator."
• Inline ED/ES Phases Recognition: Not explicitly stated; "gold standard phase indices" are referenced, implying a pre-defined or established method without detailing a multi-reader adjudication process.
• Inline MOCO: "Finally, all ground truth was evaluated by three licensed physicians with U.S. credentials." This suggests an adjudication or confirmation process, but the specific method (e.g., 2+1, consensus) is not detailed beyond "evaluated by."

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 comparative effectiveness study was explicitly described to evaluate human reader improvement with AI assistance. The described studies focus on the standalone performance of the algorithms or a qualitative assessment of images by radiologists for diagnostic quality.

6. If a Standalone (i.e., algorithm only without human-in-the-loop performance) Was Done

• Yes, standalone performance was done for all listed algorithms.
  • ACS: Evaluated quantitatively (SNR, Resolution, Contrast, Uniformity, Structure Measurement) and then qualitatively by radiologists. The quantitative metrics are standalone.
  • SparkCo: Quantitative metrics (Detection Accuracy, PSNR) and qualitative assessment by an experienced evaluator. The quantitative metrics are standalone.
  • Inline ED/ES Phases Recognition: Evaluated quantitatively as the error between algorithmic output and gold standard. This is a standalone performance metric.
  • Inline MOCO: Evaluated using the Dice coefficient, which is a standalone quantitative metric comparing algorithm output to ground truth.

7. The Type of Ground Truth Used

• AI-assisted Compressed Sensing (ACS):
  • Quantitative: Fully-sampled k-space data transformed to image space.
  • Clinical: Radiologist evaluation ("American Board of Radiologists certificated physicians").
• SparkCo:
  • Spark Detection Module: Location of spark points (ground truth for simulated data).
  • Spark Correction Module: Visual assessment by "one experienced evaluator."
• Inline ED/ES Phases Recognition: "Gold standard phase indices" (method for establishing this gold standard is not detailed, but implies expert-derived or a highly accurate reference).
• Inline MOCO: Left ventricular myocardium segmentation annotated by a "well-trained annotator" and "evaluated by three licensed physicians with U.S. credentials." This is an expert consensus/pathology-like ground truth.

8. The Sample Size for the Training Set

• AI-assisted Compressed Sensing (ACS): 1,262,912 samples (from a variety of anatomies, image contrasts, and acceleration factors).
• SparkCo: 24,866 spark slices (generated from 61 spark-free cases from 10 volunteers).
• Inline ED/ES Phases Recognition: Not explicitly provided, but stated to be "independent of the data used to test the algorithm."
• Inline MOCO: Not explicitly provided, but stated to be "independent of the data used to test the algorithm."

9. How the Ground Truth for the Training Set Was Established

• AI-assisted Compressed Sensing (ACS): Fully-sampled k-space data were collected and transformed to image space as the ground-truth. All data were manually quality controlled.
• SparkCo: "The training dataset for the AI module in SparkCo was generated by simulating spark artifacts from spark-free raw data... a total of 24,866 spark slices, along with the corresponding ground truth (i.e., the location of spark points), were generated for training." This indicates a hybrid approach using real spark-free data to simulate and generate the ground truth for spark locations.
• Inline ED/ES Phases Recognition: Not explicitly provided.
• Inline MOCO: Not explicitly provided.

Ask a specific question about this device

The uMR Omega system is indicated for use as a magnetic resonance diagnostic device (MRDD) that produces sagittal, transverse, coronal, and oblique cross sectional images, and spectroscopic images, and that display internal anatomical structure and/or function of the head, body and extremities.

These images and the physical parameters derived from the images when interpreted by a trained physician yield information that may assist the diagnosis. Contrast agents may be used depending on the region of interest of the scan.

The uMR Omega is a 3.0T superconducting magnetic resonance diagnostic device with a 75cm size patient bore. It consists of components such as magnet, RF power amplifier, RF coils, gradient power amplifier, gradient coils, patient table, spectrometer, computer, equipment cabinets, power distribution system, internal communication system, and vital signal module etc. The uMR Omega Magnetic Resonance Diagnostic Device is designed to conform to NEMA and DICOM standards.

This traditional 510(k) is to request modifications for the cleared uMR Omega(K240540). The modifications performed on the uMR Omega in this submission are due to the following changes that include:

1. Addition of RF coils and corresponding accessories: Breast Coil - 12, Biopsy Configuration, Head Coil - 16, Positioning Couch-top, Coil Support, Tx/Rx Head Coil.

2. Modification of the mmw component name: from mmw100 to mmw101.

3. Modification of the dimensions of detachable table: from width 826mm, height 880mm, length 2578mm to width 810mm, height 880mm, length 2505mm.

4. Addition and modification of pulse sequences:

   • a) New sequences: gre_pass, gre_mtp, epi_dti_msh, gre_fsp_c(3D LGE).

   • b) Added Associated options for certain sequences: fse(MicroView), fse_mx(MicroView), gre(Output phase image), gre_swi(QSM),
     gre_fsp_c(DB/GB PSIR), gre_bssfp(TI Scout), gre_bssfp_ucs(Real Time Cine), epi_dwi(IVIM), epi_dti(DSI, DKI).

   • c) Added Additional accessory equipment required for certain sequences: gre_bssfp (Virtual ECG Trigger).

   • d) Added applicable body parts: epi_dwi_msh, gre_fine, fse_mx.

5. Addition of imaging processing methods: Inline Cardiac function, Inline ECV, Inline MRS, Inline MOCO and MTP.

6. Addition of workflow features: EasyFACT, TI Scout, EasyCrop, ImageGuard, MoCap and Breast Biopsy.

7. Addition of image reconstruction methods: SparkCo.

8. Modification of function: uVision (add Body Part Recognization), EasyScan(add applicable body parts).

The modification does not affect the intended use or alter the fundamental scientific technology of the device.

The provided text describes modifications to an existing MR diagnostic device (uMR Omega) and performs non-clinical testing to demonstrate substantial equivalence to predicate devices. It specifically details the acceptance criteria and study results for two components: SparkCo (an AI algorithm for spark artifact correction) and Inline ECV (an image processing method for extracellular volume fraction calculation).

Here's a breakdown of the requested information:

Acceptance Criteria and Device Performance for uMR Omega

1. Table of Acceptance Criteria and Reported Device Performance

For SparkCo (Spark artifact Correction):

2. Based on the real-world spark dataset, evaluating the image quality improvement between the spark-corrected images and spark images by one experienced evaluator. | 1. The average PSNR of spark-corrected images needs to be higher than the spark images.
3. Spark artifacts need to be reduced or corrected after enabling SparkCo. | 1. The average PSNR of spark-corrected images is 1.6 higher than the spark images.
4. The images with spark artifacts were successfully corrected after enabling the SparkCo. |

For Inline ECV (Extracellular Volume Fraction):

The criteria is as follows:
• Satisfied (S): the segmentation myocardial boundary adheres to the myocardial boundary and blood pool ROI is within the blood pool excluding the papillary muscles.
• Acceptable (A): These are small missing or redundant areas in the myocardial segmentation but not obviously and the blood pool ROI is within the blood pool excluding the papillary muscles.
• Fail (F): The myocardial mask does not adhere to the myocardial boundary or the blood pool ROI is not within the blood pool, or the blood pool ROI contains papillary muscles. | The segmentation algorithm performed as expected in different subgroups.
Total satisfaction Rate (S): 100% for all monitored demographic and acquisition subgroups, which means no failure cases (F) or acceptable cases (A) were reported. |

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

For SparkCo:

• Test Set Sample Size:
  • Simulated Spark Testing Dataset: 159 spark slices.
  • Real-world Spark Testing Dataset: 59 cases from 15 patients.
• Data Provenance:
  • Simulated Spark Testing Dataset: Generated by simulating spark artifacts from spark-free raw data (61 cases from 10 volunteers, various body parts and MRI sequences).
  • Real-world Spark Testing Dataset: Acquired using uMR 1.5T and uMR 3T scanners, covering representative clinical protocols (T1, T2, PD with/without fat saturation) from 15 patients. The ethnicity for this dataset is 100% Asian, and the data originates from an unspecified location, but given the manufacturer's location (Shanghai, China), it is highly likely to be China. This appears to be retrospective as patients data is mentioned.

For Inline ECV:

• Test Set Sample Size: 90 images from 28 patients.
• Data Provenance: The distribution table shows data from patients with magnetic field strengths of 1.5T and 3T. Ethn_icity is broken down into "Asia" (17 patients) and "USA" (11 patients). This indicates a combined dataset potentially from multiple geographical locations, and appears to be retrospective.

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

For SparkCo:

• Spark detection accuracy: The ground truth for spark detection accuracy was established by comparing to "ground-truth" spark locations, which were generated as part of the simulation process for the training data and likely also for evaluating the testing set during the simulation step. For the real-world dataset, the document mentions "comparing the spark detection results with the ground-truth" implying an existing ground truth, but doesn't specify how it was established or how many experts were involved.
• Spark correction performance: "One experienced evaluator" was used for subjective evaluation of image quality improvement on the real-world spark dataset. No specific qualifications are provided for this evaluator beyond "experienced".

For Inline ECV:

• The document states, "The segmentation result of each case was obtained with the algorithm, and the segmentation mask was evaluated with the following criteria." It does not explicitly mention human experts establishing a distinct "ground truth" for each segmentation mask for the purpose of the acceptance criteria. Instead, the evaluation seems to be a subjective assessment against predefined criteria. No number of experts or qualifications are provided.

4. Adjudication Method for the Test Set

For SparkCo:

• For spark detection accuracy, the comparison was against a presumed inherent "ground-truth" (likely derived from the simulation process).
• For spark correction performance, a single "experienced evaluator" made the subjective assessment, implying no adjudication method (e.g., 2+1, 3+1) was explicitly used among multiple experts.

For Inline ECV:

• The evaluation was a "subjective evaluation method" against specific criteria. No information about multiple evaluators or an adjudication method is provided. It implies a single evaluator or an internal consensus without formal adjudication.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was Done

• No, a Multi-Reader Multi-Case (MRMC) comparative effectiveness study comparing human readers with and without AI assistance was not explicitly mentioned for either SparkCo or Inline ECV. The studies were focused on the standalone performance of the algorithms.

6. If a Standalone (i.e., algorithm only without human-in-the-loop performance) was Done

• Yes, for both SparkCo and Inline ECV, the studies described are standalone algorithm performance evaluations.
  • SparkCo focused on the algorithm's ability to detect and correct spark artifacts (objective metrics like PSNR and subjective assessment by one evaluator).
  • Inline ECV focused on the algorithm's segmentation accuracy (subjective evaluation of segmentation masks against criteria).

7. The Type of Ground Truth Used

For SparkCo:

• Spark detection accuracy: Ground truth was generated by simulating spark artifacts from spark-free raw data, implying a simulated/synthetic ground truth for training and a comparison against this for testing. For real-world data, the "ground-truth" for detection is implied but not explicitly detailed how it was established.
• Spark correction performance: For PSNR, the "ground truth" for comparison is the original spark images. For subjective evaluation, it's against the "spark images" and the expectation of correction, suggesting human expert judgment (by one evaluator) rather than a pre-established clinical ground truth for each case.

For Inline ECV:

• The ground truth for Inline ECV appears to be a subjective expert assessment (though the number of experts is not specified) of the algorithm's automatically generated segmentation masks against predefined "Satisfied," "Acceptable," and "Fail" criteria. It is not an independent, pre-established ground truth like pathology or outcomes data.

8. The Sample Size for the Training Set

For SparkCo:

• Training dataset for the AI module: 61 cases from 10 volunteers. From this, a total of 24,866 spark slices along with corresponding "ground truth" (location of spark points) were generated for training.

For Inline ECV:

• The document states, "The training data used for the training of the cardiac ventricular segmentation algorithm is independent of the data used to test the algorithm." However, the sample size for the training set itself is not explicitly provided in the given text.

9. How the Ground Truth for the Training Set Was Established

For SparkCo:

• The ground truth for the SparkCo training set was established by simulating spark artifacts from spark-free raw data. This simulation process directly provided the "location of spark points" as the ground truth.

For Inline ECV:

• The document mentions that the training data is independent of the test data, but it does not describe how the ground truth for the training set of the Inline ECV algorithm was established.

Ask a specific question about this device