JF’s unique cloud-based R&D platform with high quality and productivity:

Completely cloud-based online model iteration: efficient reuse of production data, standard image labeling, and annotation, real-time online feedback from in-house radiologists, continuous deployment of AI model, big data analysis. Using neural architecture search, a technique Google developed to automatically design neural networks with neural networks, JF’s AI team designs a novel deep convolutional backbone and loss function to model the specific image features of X-rays. The AI model can characterize different types of thorax diseases with annotations.

JF CXR-1

The AI algorithm analyzes the image and outputs a probability score of TB and highlight suspected lesions with heatmaps to assist doctors for TB detection.

Tuberculosis heatmap

Training

100
Normal
34k
Radiologist
confirmed TB
20k
Other
abnormalities

Testing

4981
Normal
2290
Bacteriologically
confirmed TB
2090
Other
abnormalities

Results

0.96
AUC
95%
Sensitivity
70%
Specificity
96%
Radiologist
consistency
* All data were curated by radiologists with over 20 year‘s working experience

JF CXR-2

The AI algorithm uses cutting-edge deep learning technology to detect multiple major thorax diseases simultaneously on chest and bone X-rays, including tuberculosis, pneumonia, lung nodule/mass, pleural diseases, etc. For each disease, the AI solution generates annotation, description of lesion signs, image evaluation, AI-assisted diagnosis report, etc.

Tuberculosis
Pneumonia
Lung nodule/mass
Pleural diseases

Testing

  • - Test data from multiple provinces covering Eastern, Middle and Western China
  • - Randomly selected 1000 chest X-ray images from township hospitals
  • - Three senior radiologists independently read the chest X-rays and the result from majority is the Ground Truth
  • - Multi-thorax diseases detection all achieves more than 90% accuracy

Results

97%
Tuberculosis
90%
Pneumonia
90%
Lung nodule/mass
99%
Pleura/chest

To meet the needs of primary care settings, JF Healthcare has established a technological platform to address the “pain points” along the entire medical imaging process, which includes mainly four modules: image quality control, technician guidance, intelligent image evaluation, and big data analysis.

Big Data Analysis and Visualization Monitoring Platform

The big data analysis function of the system is based on a large amount of business data generated by daily work processes, and allows management to see trends to improve quality, productivity, or even to spot a regional TB outbreak in order to prevent local epidemics before they begin.

JF Smart Cloud
Telemedicine Platform

Diagnostic evaluation by AI algorithm based on cloud computing. Big data storage with high security; no need for film printing, permanent data storage and analysis, ready-to-use, high processing efficiency.