top of page
Last Updated:
28 September 2021
ImagInfer is a deep learning-based inferring solution for tuberculosis (TB) detection and can be further trained with data from the client to recognize other conditions, such as pneumonia and COVID-19 on chest X-rays. This solution can have multiple use cases. Where a radiologist is not available (for instance in rural areas), inference can happen with high accuracy. In big hospitals in developed countries where radiologists may have limited time, the system can reprioritize X-rays for reporting instead of “first in first out”.
New models for new conditions can also be created with the training data provided.
Certification
Pending. Undergoing validation at State TB Offices in India. CE certification planned
Development Stage
Validation
Deployment
Hybrid (offline analysis of the image but online on demand synchronization with legacy systems or cloud storage)
Intended Age Group
8+ years
Target Setting
Primary health centres, general hospital (above primary level), teleradiology companies, government/public sector, e.g. national TB programme, private sector
Current Market
India, Africa
Input
»» Chest X-ray image format: JPEG, PNG, DICOM, Tiff, PDF
»» Chest X-ray type: posterior-anterior chest X-ray, anterior-posterior chest X-ray, lateral chest X-ray, and any X-rays as long as they are used for training the model
Output
»» Dichotomous output indicating whether each abnormality is present or absent
»» Dichotomous output only indicating whether TB is likely present or likely absent
»» No score is provided
1/1
Hardware
This is dependent on the volume concurrency and expected speed. In general, a 24 GPU machine is required for training and a 12 GPU machine is required for inference.
Validation
No validation is required for the X-ray machine. The quality of the output from the X-ray machine should be good and if the model is trained using the output from the same X-ray machine, the accuracy will be very high.
Hardware
This is dependent on the volume concurrency and expected speed. In general, a 24 GPU machine is required for training and a 12 GPU machine is required for inference.
Server
Hybrid deployment model, which can be set up anywhere. On premise deployment can be done as well as a cloud deployment.
Integration
Integrations and other customizations may be possible on request. An appropriate quote will be provided.
Software
-
Processing Time
6 seconds
Data Sharing & Privacy
»» Server location (for online product): A hybrid deployment model, can be set up anywhere. On-premise deployment can be done, as well as a cloud deployment.
»» Data are not shared with the manufacturer
»» There is no option to deidentify data.
»» The image data and not the corresponding metadata present in the file is used. For example, if a DICOM file is given as input, then only the image data from that and not the metadata, that constitutes the information about the patient, is utilized.
Software Updates
»» Depends on any new features added. In general, a 6-month window is kept for upgrades.
»» Updates will normally be free if the client is under an annual maintenance contract. If some completely new modules are added, then the pricing will be modified accordingly as per the module.
Product Development Method
Supervised deep learning (CNN, RNN)
Training
670 adult chest X-rays to train and create a model for any condition. Public datasets from China.
Reference Standard
Mostly human readers
Publications
Peer-reviewed publications are not yet available.
This website works best with browsers other than Internet Explorer.
bottom of page