Version 2



Company HQ: 

Delaware, USA

Last Updated:

April 14th, 2020

DxTB highlights regions of a chest X-ray that have specific abnormalities associated with tuberculosis (TB) and helps radiologists to prioritize and reduce their workload. DxTB is an adaptive system that evolves over time based on every new X-ray read in the field. DxTB also automatically generates Radiological Society of North America (RSNA) standard radiology reports.


CE and FDA mark pending (expected Q2 2021)

Development Stage

On the market


Online & offline (with intermittent internet connection) with local edge computing device.

Intended Age Group

14+ years

Target Setting

Primary health centers, general hospital (above primary level), teleradiology companies, government/public sector, e.g. national TB program, and private sector. An automated solution or a teleradiology-backed AI diagnosis and report is provided depending on the situation.

Current Market

India, Asia Pacific, and Africa. Future market (after CE/FDA certification): the USA, Europe and Japan


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

»» Chest X-ray image format: PNG, DICOM

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

»» Other requirements: requires DICOM-compatible images


Structured report including: 

»» Heat map, 

»» Probability score for TB, 

»» Probability score for the following pulmonary abnormalities: abscess, airfluid level, atelectasis, blunted costophrenic angle, bronchiectasis, calcification, cavity, consolidation, fibrosis, interstitial markings, loculated pleural effusion, lymphadenopathy, mass, nodule, opacity, pleural effusion, prominence in hilar region, pneumothorax, tracheal shift, granuloma, calcified pleural plaques. 

»» Location of certain abnormalities. 

»» Binary outputs using client-specific threshold scores.



Server requirements for predictions are the same as the hardware requirements.


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


Linux and some open source deep learning software

Processing Time

2 seconds


I5/I7 processor or equivalent, 6GB RAM, 1TB hard-drive


Data will be required from new clients to validate the distribution similarity with training data. If the new data is not similar to the training dataset, transfer learning may be required.

Data Sharing & Privacy

»» Server Location: local installation either in cloud or on-premises possible

»» X-ray and if available patient’s clinical data are uploaded to a cloud platform. If the client provides consent, data is shared with the developer in an anonymized form to improve the models. HIPPA guidelines are followed.

»» There is an option to de-identify data

Software Updates

»» Monthly software updates

»» Software upgrades are included in the license price

»» Extra costs: none


»» Volume-based pricing models are available

»» Please contact company for quote: Aniruddha Pant at


I5/I7 processor or equivalent, 6GB RAM, 1TB hard-drive

Product Development Method

Unsupervised deep learning (AEs/SAE, RBMs, DBNs) and supervised deep learning (CNN, RNN)


The product was trained on 500 000 chest X-rays from India, the USA, China, and Malaysia

Reference Standard

Human reader. Culture and GeneXpert based training will be conducted in the future


»» Kulkarni V, Kulkarni M, Pant A. Survey of personalization techniques for federated learning. arXiv preprint arXiv:2003.08673. 2020.

Does this page require updates? Send us a message:

Designed by Tasneem Naheyan

This website works best with browsers other than Internet Explorer.