Last Updated:
20 8 2026
VitaForesee is AI-powered CAD software for chest X-ray TB screening and triage, designed for large-scale use in primary care and high-burden settings. It can be deployed with ultra-portable digital radiography for point-of-care screening.
For people aged 15+, the system analyzes CXR images in seconds and provides a TB probability score, heatmap, abnormality localization and structured report. It detects TB and 18 other chest abnormalities.
Multicentre clinical validation reported 92% TB sensitivity, 95.5% specificity and 92.49% agreement with physician interpretation. The software supports DICOM/PACS integration, desktop alerts and exportable diagnostic reports.
Registered as a medical device in China, the product supports local/offline deployment, automated de-identification and SM4 encryption. Intended settings include community screening, mobile outreach and facility-based triage.
Certification
Version V1: China Medical Device Registration Certificate (NMPA). Authority: Zhejiang Medical Products Administration. Registered product: Medical Image Processing Software.
Development Stage
On the Market
Deployment
Online & Offline
Intended Age Group
15+ years
Target Setting
Primary health centres; secondary and tertiary hospitals; teleradiology providers; government/public-sector programmes, including National TB Programmes; and private-sector providers.
Current Market
China, including primary care, health examination and public-health screening settings.
Input
Can be used to read images from any digital chest X-ray machine and model
Chest X-ray image format: DICOM
Chest X-ray type: Posterioranterior (PA) chest X-ray, Anteroposterior (AP) CXR
Output
Output includes:
Heatmap
Dichotomous output indicating whether TB is likely present or absent
Dichotomous output indicating whether each abnormality is likely present or absent
Probability score for TB
Probability score for each abnormality
Location of each abnormality
Default threshold for TB is 0.5. The threshold can be adjusted from 0 to 1 to match the screening population and programme objective, allowing users to balance case detection and referral volume.
The system automatically generates structured reports with annotated images, AI scores and diagnostic findings. Reports can be viewed as text or image-and-text, exported as PDF, or printed. Selected reports can also be batch printed.
Lung abnormalities included in the TB Score: Fibrosis, Loculated pleural effusion, Pneumothorax, Pleural thickening
Additional findings reported by the product: Atelectasis, Blunted costophrenic angle, Consolidation, Fibrosis, Mass, Nodule, Pleural effusion, Pneumothorax, Pleural thickening, Fracture, Diaphragmatic hernia, Emphysema, Pulmonary edema
Hardware
Minimum requirements:
Intel Core i5, 8 GB RAM, 500 GB storage, Windows 11 64-bit. Configuration can be scaled based on daily screening volume and concurrent workload.
Server
Supports mainstream VPS providers, private cloud and customer-hosted on-premise servers. Server sizing depends on expected daily volume, concurrent workload and data-retention period. VitaForesee provides deployment assessment and configuration guidance.
Integration with X-ray Systems
Integration with PACS and Legacy Systems
The software can integrated with third-party X-ray systems. The software receives DICOM 3.0 CXR images via DICOM Storage SCU/SCP and AET connections, including concurrent feeds from multiple X-ray systems. Compatible examples include Siemens, GE, Philips, Carestream, United Imaging, WDM, Neusoft and Angell.
Integrates with DICOM 3.0 PACS via Storage SCU/SCP and AET. Images can be routed automatically to AI, with annotations and structured reports returned to PACS. A lightweight desktop-assistant option supports rapid integration without changes to the PACS core.
Software
Browser-based B/S architecture. Compatible with mainstream browsers including Chrome and Edge. No dedicated image-viewing client is required.
Processing Time
Typically 1-3 seconds per standard DICOM CXR for AI analysis, abnormality detection and structured-report generation. Time may vary with file size, deployment configuration and network conditions.
Data Sharing & Privacy
Patient images and clinical data do not need to be shared with the developer. On-premise and standalone offline deployment keep data local. Automated de-identification is available, and transmission/storage support SM4 encryption. Data sharing can be configured to customer policy.
Software Updates
Functional updates and bug fixes are typically released about every 15 days; AI model updates are provided as needed. Updates are scheduled during off-peak periods and may briefly interrupt service. Local data are retained, and existing PACS/DR integration rules are maintained.
Price
Pricing is flexible and depends on deployment model and screening volume. Available options include:
1. Annual licence: software licence plus annual maintenance, with unlimited reads. Includes version updates, remote technical support and interface maintenance. Available for on-premise or cloud deployment.
2. Pay-per-read: a negotiated fee per CXR, with no high upfront licence fee. Annual maintenance may apply and covers system support and fault resolution. Suitable for projects or gradually increasing screening volumes.
3. Upfront software and on-site installation fees may be waived depending on the package. Annual maintenance may cover updates, technical integration and security checks. Pay-per-read packages have no additional per-image surcharge beyond the agreed read fee.
4. Custom/hybrid packages are available based on screening volume and deployment model (offline workstation, on-premise or cloud). Contact 4058948@qq.com for a formal quotation.
Product Development Method
The model combines Transformer + YOLO with feature-attention, scale-aware super-resolution, omni-supervised and weakly supervised learning for multi-abnormality detection and localization. Related methods have been published in MICCAI, Radiology: Artificial Intelligence and IEEE TMI.
Training
The AI model was trained on approximately 800,000 standardized chest X-rays.
More detailed information on data sources and population characteristics can be provided for project evaluation.
Reference Standard
Human radiograph interpretation, culture, smear microscopy, GeneXpert and CT.
Publications
[1] Luo, L., Chen, H., Zhou, Y., Lin, H., Heng, P.A. (2021). OXnet: Deep Omni-Supervised Thoracic Disease Detection from Chest X-Rays. In: de Bruijne, M., et al. Medical Image Computing and Computer Assisted Intervention - MICCAI 2021. Lecture Notes in Computer Science, vol. 12902. Springer, Cham. https://doi.org/10.1007/978-3-030-87196-3_50
[2] Luo, L., Chen, H., Xiao, Y., Zhou, Y., Wang, X., Vardhanabhuti, V., Wu, M., Han, C., Liu, Z., Fang, X.H.B., Tsougenis, E., Lin, H., Heng, P.A. (2022). Rethinking Annotation Granularity for Overcoming Shortcuts in Deep Learning-based Radiograph Diagnosis: A Multicenter Study. Radiology: Artificial Intelligence, 4(5): e210299. https://doi.org/10.1148/ryai.210299
[3] Chai, Z., Luo, L., Lin, H., Heng, P.A., Chen, H. (2024). Deep Omni-Supervised Learning for Rib Fracture Detection from Chest Radiology Images. IEEE Transactions on Medical Imaging, 43(5): 1972-1982. https://doi.org/10.1109/TMI.2024.3353248
[4] Chen, C., Dou, Q., Chen, H., Heng, P.A. (2018). Semantic-Aware Generative Adversarial Nets for Unsupervised Domain Adaptation in Chest X-ray Segmentation. arXiv preprint arXiv:1806.00600. https://doi.org/10.48550/arXiv.1806.00600
This website works best with browsers other than Internet Explorer.