Search Results

The system is a PET/CT system designed for providing anatomical and functional images. The PET provides the distribution of specific radiopharmaceuticals. CT provides diagnostic tomographic anatomical information as well as photon attenuation information for the scanned region. PET and CT scans can be performed separately. The system is intended for assessing metabolic (molecular) and physiologic functions in various parts of the body. When used with radiopharmaceuticals approved by the regulatory authority in the country of use, the system generates images depicting the distribution of these radiopharmaceuticals. The images produced by the system are intended for analysis and interpretation by qualified medical professionals. They can serve as an aid in detection, localization, evaluation, diagnosis, staging, re-staging, monitoring, and/or follow-up of abnormalities, lesions, tumors, inflammation, infection, organ function, disorders, and/or diseases, in several clinical areas such as oncology, cardiology, neurology, infection and inflammation. The images produced by the system can also be used by the physician to aid in radiotherapy treatment planning and interventional radiology procedures.

The CT system can be used for low dose CT lung cancer screening for the early detection of lung nodules that may represent cancer. The screening must be performed within the established inclusion criteria of programs / protocols that have been approved and published by either a governmental body or professional medical society.

Device Description

The proposed device uMI Panvivo combines a 295/235/534/712 mm axial field of view (FOV) PET and 160-slice CT system to provide high quality functional and anatomical images, fast PET/CT imaging and better patient experience. The system includes PET system, CT system, patient table, power distribution unit, control and reconstruction system (host, monitor, and reconstruction computer, system software, reconstruction software), vital signal module and other accessories.

The uMI Panvivo has been previously cleared by FDA via K251839.The main modifications performed on the uMI Panvivo (K251839) in this submission are the addition of two new models. The previous uMI Panvivo(K251839) is designed with scalable PET rings; uMI Panvivo ES is scaling to 180 PET rings and uMI Panvivo EX is scaling to 240 PET rings, compares to the uMI Panvivo 100 PET rings and uMI Panvivo S 80 PET rings.

AI/ML Overview

Here's a breakdown of the acceptance criteria and study details for the AI-powered features of the uMI Panvivo PET/CT Systems, based on the provided FDA 510(k) clearance letter:

Overview of AI Features and their Studies for uMI Panvivo PET/CT Systems

This document describes the acceptance criteria and study methodologies for four AI-powered features integrated into the uMI Panvivo PET/CT Systems: DeepMAC, uExcel DPR, OncoFocus, and NeuroFocus.Brain, as well as AIEFOV.

1. DeepMAC (Metal Artifact Reduction)

DeepMAC is an image post-processing technology that uses pre-trained neural networks to reduce metal artifacts and improve image quality in CT scans.

Acceptance Criteria and Reported Device Performance for DeepMAC:

Evaluation Item	Criteria	Reported Device Performance
Quantitative evaluation	For PMMA phantom data, the average CT value in the affected area of the metal substance and the same area of the control image before and after DeepMAC was compared. After using DeepMAC, the difference between the average CT value in the affected area of the metal substance and the same area of the control image does not exceed 10HU.	Pass
Clinical Evaluation	Radiologists verified that DeepMAC effectively corrects metal artifacts and improves tissue interpretability.	Verified

Study Details for DeepMAC:

Sample size for the test set: 20 human subjects (clinical dataset) and PMMA phantom datasets.
Data provenance: Collected from various clinical sites, different from the training data. No overlap between training and testing data.
Number of experts and qualifications (for ground truth/evaluation): Two American Board of Radiologists certified physicians.
Adjudication method for the test set: Not explicitly stated for quantitative evaluation. For clinical evaluation, radiologists independently evaluated images.
MRMC comparative effectiveness study: Not explicitly mentioned as a formal MRMC study design. However, clinical images were evaluated by radiologists to verify effectiveness.
Standalone performance: The quantitative evaluation on PMMA phantom data represents standalone algorithmic performance.
Type of ground truth: For phantom data: "control image" (presumably ideal images without artifacts). For training data: "corresponding ground truth images without metal artifacts" derived from system simulations. For clinical evaluation: expert opinion from board-certified radiologists.
Sample size for the training set: Not explicitly stated beyond "system simulations".
How ground truth for training set established: Derived from system simulations, with pairs of image data: images with metal artifacts and corresponding ground truth images without metal artifacts.

2. uExcel DPR (Deep Progressive Reconstruction for PET)

uExcel DPR is a deep learning-based PET reconstruction algorithm designed to optimize the iterative reconstruction process, reduce noise, and improve contrast by utilizing pre-trained deep neural networks.

Acceptance Criteria and Reported Device Performance for uExcel DPR:

Evaluation Item	Criteria	Reported Device Performance
NEMA IQ phantom analysis	Contrast recovery (CR), background variability (BV), and contrast-to-noise ratio (CNR) were calculated using NEMA IQ phantom data reconstructed with uExcel DPR and OSEM under acquisition conditions of 1 to 5 minutes per bed. The averaged CR, BV, and CNR of the uExcel DPR images should be superior to those of the OSEM images.	Pass (Maximum noise reduction of 47%, average SNR improvement of 131%)
Human subject evaluations	A comparative evaluation of uExcel DPR and OSEM reconstructed images was conducted through independent visual assessments and quantitative liver signal-to-noise ratio (liverSNR) analyses. uExcel DPR demonstrate superior image SNR compared to OSEM reconstruction across various counting conditions.	Pass (Superior image SNR across diverse counting conditions)
Clinical Evaluation	All images were adequate for clinical diagnosis. Images reconstructed using the uExcel DPR algorithm exhibited lower noise, improved contrast, and greater sharpness compared to those reconstructed with the OSEM algorithm.	Verified

Study Details for uExcel DPR:

Sample size for the test set: 8 human subjects (clinical dataset) and two NEMA IQ phantom datasets.
Data provenance: Collected from uMI Panvivo EX and uMI Panvivo ES systems. Testing data are entirely independent from the training data, collected using different types of PET/CT scanners. Asian ethnicity (100%).
Number of experts and qualifications (for ground truth/evaluation): Two American board-certified nuclear medicine physicians.
Adjudication method for the test set: Blind comparison between images reconstructed using uExcel DPR and OSEM algorithms. Physicians evaluated images independently.
MRMC comparative effectiveness study: Yes, a blind comparison was conducted, showing images with uExcel DPR had lower noise, improved contrast, and greater sharpness compared to OSEM. The effect size, though not quantified by a specific metric like AUC improvement, states a "maximum noise reduction of 47% and an average SNR improvement of 131%" in phantom analysis, and "superior image SNR" in human subjects.
Standalone performance: Yes, NEMA IQ phantom analysis directly assesses algorithmic performance.
Type of ground truth: For training: "Full-sampled data" serves as ground truth, while "corresponding down-sampled data" (created with varying factors) acts as training input. For validation: NEMA IQ phantom standards; OSEM reconstruction images for comparison; expert opinion from board-certified nuclear medicine physicians.
Sample size for the training set: Not explicitly stated, sourced from uEXPLORER and uMI Panorama GS PET/CT systems.
How ground truth for training set established: Full-sampled data from long-axis datasets (from uEXPLORER and uMI Panorama GS PET/CT systems) served as the ground truth.

3. OncoFocus (Respiratory Motion Correction)

OncoFocus is a motion correction technique that uses deep learning to correct respiratory motion artifacts in PET/CT images, improving accuracy of SUV and lesion volume. It includes a body cavity segmentation network (CNN-SEG) and an attenuation map synthesis network (CNN-AC).

Acceptance Criteria and Reported Device Performance for OncoFocus:

Evaluation Item	Criteria	Reported Device Performance
Volume relative to no respiratory motion correction (∆Volume)	Calculating the OncoFocus volume change relative to no respiratory motion correction images. The ∆Volume value is less than 0%.	Pass (average lesion volume is smaller)
Maximal standardized uptake value relative to no respiratory motion correction (∆SUVmax)	Calculating the SUVmax obtained from the OncoFocus with that from the corresponding non-corrected image. The ∆SUVmax value is large than 0%.	Pass (average lesion SUVmax is superior)
Clinical Evaluation	Radiologists verified that OncoFocus can reduce respiratory motion artifacts, yield higher PET/CT alignment accuracy, and enhance diagnostic confidence compared with NMC (non-motion correction) images.	Verified

Study Details for OncoFocus:

Sample size for the test set: 13 human subjects (clinical patient cases) specifically tested on uMI Panvivo EX and uMI Panvivo ES.
Data provenance: Collected from clinical scenarios, different from the training data. No overlap between training and testing data. Asian ethnicity (100%).
Number of experts and qualifications (for ground truth/evaluation): Two American Board of Radiologists-certified physicians.
Adjudication method for the test set: Not explicitly stated for quantitative metrics. For clinical evaluation, radiologists independently compared OncoFocus images and NMC images.
MRMC comparative effectiveness study: Not explicitly mentioned as a formal MRMC study design. Clinical evaluation involved comparison by radiologists, showing enhancement of diagnostic confidence.
Standalone performance: Yes, quantitative measurements like ∆Volume and ∆SUVmax demonstrate algorithmic performance.
Type of ground truth: For training data:
- CNN-SEG: Input data are CT-derived attenuation coefficient maps; target data are body cavity region images.
- CNN-AC: Input data are non-attenuation-corrected (NAC) PET reconstruction images; target data are reference CT attenuation coefficient maps.
  For validation: NMC images for comparative quantitative analysis; expert opinion from board-certified radiologists.
Sample size for the training set: Not explicitly stated, collected from "general clinical scenarios" using UIH PET/CT systems.
How ground truth for training set established:
- CNN-SEG: Target data (body cavity region images) were established based on CT-derived attenuation coefficient maps.
- CNN-AC: Target data (reference CT attenuation coefficient maps) were established using NAC PET reconstruction images.

4. NeuroFocus.Brain (Head Artifact Elimination in Brain PET)

NeuroFocus.Brain is a motion management technology that incorporates AI to eliminate head artifacts in brain PET imaging, automatically detecting motion and selecting motion-free data for reconstruction. It includes a brain segmentation network (CNN-SEG) and a CNN-based attenuation map synthesis network (CNN-AC).

Acceptance Criteria and Reported Device Performance for NeuroFocus.Brain:

Evaluation Item	Evaluation Method	Criteria	Reported Device Performance
Quantitative evaluation	Calculate ∆SUVmean in the high-uptake region for two MCS (Monte Carlo-Simulated) cases: one with motion introduced during simulation and reconstructed using NeuroFocus.Brain, and one stationary reconstructed without NeuroFocus.Brain.	The ∆SUVmean value is less than 10%.	Pass (effectively corrects quantitative reduction)
	Calculate ∆SUVmean in the high-uptake region of the prefrontal cortex, relative to reconstruction without NeuroFocus.Brain for the same clinical scan with head motion.	The ∆SUVmean value is large than 0%.	Pass (significantly improves quantitative accuracy)
Clinical Evaluation	Radiologists verified that NeuroFocus.Brain can reduce head motion artifacts and improve image quality, thereby enhancing diagnostic confidence compared with images reconstructed without NeuroFocus.Brain.		Verified

Study Details for NeuroFocus.Brain:

Sample size for the test set: One Monte Carlo-simulated brain phantom case (with motion and stationary scenarios) and 7 human subjects (clinical volunteer cases with notable head motion artifacts).
Data provenance: For clinical cases: retrospecitively identified. Testing datasets for the networks were collected on a scanner different from the one used for the training data, ensuring complete separation. Asian ethnicity (100%).
Number of experts and qualifications (for ground truth/evaluation): Two American Board of Radiologists-certified physicians.
Adjudication method for the test set: Not explicitly stated for quantitative metrics. For clinical evaluation, radiologists independently compared reconstructed images with and without NeuroFocus.Brain.
MRMC comparative effectiveness study: Not explicitly mentioned as a formal MRMC study design. Clinical evaluation involved comparison by radiologists, showing enhancement of diagnostic confidence.
Standalone performance: Yes, quantitative measurements on simulated phantom and clinical cases assess algorithmic performance.
Type of ground truth: For training data:
- CNN-SEG: Input data are CT-derived attenuation coefficient maps; target data are brain region images.
- CNN-AC: Input data are non-attenuation-corrected (NAC) PET reconstruction images; target data are reference CT attenuation coefficient maps.
  For validation: Monte Carlo simulated stationary scans (ideal without motion); reconstruction without NeuroFocus.Brain for clinical comparison; expert opinion from board-certified radiologists.
Sample size for the training set: Not explicitly stated, collected from "general clinical scenarios" using UIH PET/CT systems.
How ground truth for training set established:
- CNN-SEG: Target data (brain region images) were established based on CT-derived attenuation coefficient maps.
- CNN-AC: Target data (reference CT attenuation coefficient maps) were established using NAC PET reconstruction images.

5. AIEFOV (Artificial Intelligence Extended Field of View)

AIEFOV aims to improve the accuracy of CT values, and the accuracy and uniformity of PET image SUV by performing attenuation correction with CT generated by the AIEFOV algorithm, especially when the scanned object exceeds the CT field of view.

Acceptance Criteria and Reported Device Performance for AIEFOV:

| Evaluation Item | Criteria | Reported Device Performance |
| :--------------------- | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------0-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :--------------------------------------------------------- |
| Quantitative evaluation | When scanned object exceeds CT field of view: AI EFOV shall improve the accuracy of CT value, and improve the accuracy and uniformity of PET image SUV by performing attenuation correction with CT generated with AIEFOV algorithm. Compared to the ground truth, the uniformity and SUV deviation of PET image obtained by using AIEFOV for attenuation correction should be less than 5%. When scanned object does not exceed CT field of view: AI EFOV shall have consistent CT value, and PET image SUV by performing attenuation correction with CT generated with AIEFOV algorithm. | Pass (improves SUV accuracy when object exceeds CT-FOV; consistent SUV when within CT-FOV) |
| Clinical Evaluation | AI EFOV has the potential to enhance homogeneity and reduce image artifacts. | Verified |

Study Details for AIEFOV:

Sample size for the test set: Bench tests included water phantom scans. Clinical evaluation included 6740 images from 4 patients (Table 10).
Data provenance: For clinical cases, collected from uMI Panvivo EX/ES. Testing datasets were collected from various clinical sites and were different from the training data. No overlap between training and testing data. Asian ethnicity (100%).
Number of experts and qualifications (for ground truth/evaluation): Two American Board qualified clinical experts for blind comparison.
Adjudication method for the test set: Blind comparison by two experts for image artifacts, homogeneity, and diagnostic confidence.
MRMC comparative effectiveness study: Not explicitly mentioned as a formal MRMC study. Clinical evaluation involved blind comparison by experts.
Standalone performance: Yes, performance bench tests on water phantoms and quantitative SUV deviation measurements represent algorithmic performance.
Type of ground truth:
- For phantom study: the "ground truth" for SUV deviation and uniformity is implied to be a reference value from an ideal (untruncated) scan.
- For training: "simulated gold standard" consists of images free from truncation artifacts, derived from system simulations based on the same patient.
- For clinical evaluation: expert opinion from board-qualified clinical experts.
Sample size for the training set: Not explicitly stated, collected from "clinical data with different patient body sizes and different scanning positions."
How ground truth for training set established: "Simulated gold standard" (images free from truncation artifacts) was used as the network output, generated by reconstructing from data where both sides of the detector had not been truncated, contrasting with the input (images reconstructed with truncation artifacts). manually quality-controlled.

Ask a Question

Ask a specific question about this device

Page 1 of 1