JF CXR-2 is a screening and triaging tool to help clinicians identify abnormalities on chest X-rays. JF CXR-2 can also be used as a prioritization tool for radiologists and teleradiology companies.

Certification

Pending, China NMPA tier-3 approval

Development Stage

China NMPA tier-2

Deployment

Online & offline

Intended Age Group

15+ years

Target Setting

Primary health centres, teleradiology companies, government/public sector, e.g. national tuberculosis (TB) programme, private sector

Current Market

China

Input

»» Can be used to read images from any kind of chest X-ray machine (vendor neutral)

»» Chest X-ray image format: DICOM

»» Chest X-ray type: posterior-anterior chest X-ray

»» Other requirements: none

Output

Output in the form of: 

»» Heat map

»» Probability score for TB and dichotomized outcome for TB

»» Probability score for any abnormality, pneumonia, nodule

»» Pop-up alert message when TB score exceeds cut-off and a line list of all examination records which allows sorting, searching, exporting and statistical analysis

Hardware

Online: basic desktop without GPU to upload the DICOM files to the server. Recommended specs: Intel Xeon E5-2650, 8 GB memory, 500 GB disk space
Offline: Intel Xeon E5-2650, 16 GB memory, 1 TB disk space

Validation

X-ray machine validation is not needed.

Hardware

Online: basic desktop without GPU to upload the DICOM files to the server. Recommended specs: Intel Xeon E5-2650, 8 GB memory, 500 GB disk space
Offline: Intel Xeon E5-2650, 16 GB memory, 1 TB disk space

Server

A proprietary software that supports FTP/SFTP-style DICOM uploading.
Offline: none

Integration

It is possible to integrate the product with client’s legacy picture archiving and communication (PACS) system.

Software

A proprietary software that supports File Transfer Protocol (FTP)-style DICOM uploading

Processing Time

Online deployment: it takes around 1 second for the artificial intelligence (AI) algorithm to process the image and generate the result, but the time to upload the DICOM file may vary depending on the local internet environment.
Offline deployment: it takes around 5 seconds for the AI algorithm to process the image and generate the result.

Data Sharing & Privacy

»» Server location: the server is currently located on AliCloud in China. It can be deployed locally and offline

»» Upon consent, the pixel array from the DICOM file is shared when the DICOM is generated and uploaded to the company's cloud platform

»» The shared data are used for troubleshooting, monitoring product usage, training or improvement of software. The AI results are confirmed by JF Healthcare's team of radiologists. Customers are also notified of any emergent abnormalities e.g. pneumothorax/pneumoperitoneum not related to TB

»» There is an option to de-identify data

Price

»» Volume-based pricing models are available

»» Please contact company for quote: gengyu.bi@jfhealthcare.com

Software Updates

»» Quarterly to yearly software updates

»» For improving existing functions, the updates usually do not include additional fees. If new functions are added, the customer can choose whether to pay for additional functions or not. Currently, there are no cost differences between the private and public sector

»» Extra costs: none for minor updates

Product Development Method

Supervised deep learning (CNN, RNN)

Training

The product was trained on over 120 000 chest X-rays from China

Reference Standard

Human reader

Publications

»» Qin, Z.Z., Ahmed, S., Sarker, M.S. et al. Tuberculosis detection from chest x-rays for triaging in a high tuberculosis-burden setting: an evaluation of five artificial intelligence algorithms. Lancet Digital Health. 2021;3:e543–5437: 2113–31. https://doi.org/10.1016/S2589-7500(21)00116-3