Last Updated:
15th May 2023
DrAid™ is comprehensive artificial intelligence (AI) doctor assistant platform developed by VinBrain to assist medical teams and doctors in diagnosing 52 disease classes including Tuberculosis, capable to integrate seamlessly with PACS and radiology workflow. The platform assists the radiology workflow to identify abnormalities on a chest X-Ray and prioritised the radiologist's worklist.
Certification
DrAid Radiology V1 which got FDA clearance, is an AI-enabled triage and notification software designed to aid the clinical assessment of adult Chest X-ray cases with features suggestive of pneumothorax.
DrAid™ received ISO13485:2016 certification. DrAid is undergoing CE marking approval process for TB.
Development Stage
On the Market
Deployment
On-premise on dedicated hardware (server)
On-Premises virtualized (virtual machine, docker)
Cloud-based
Intended Age Group
6+ years
Target Setting
Primary health centers
General hospital
Private Healthcare org- CSR campaigns
Teleradiology companies
Medical Device companies (Digital XR)
Government / NGO Public surveillance
Current Market
Global- Vietnam & South-east asia
Input
DICOM, JPG, PNG;
Posterior-anterior CXR
Anterior-posterior CXR
Portable
Output
Heat map
Dichotomous output indicating for each abnormality whether this is present or absent.
Dichotomous output only indicating whether TB is likely present or likely absent.
Probability score for TB
Probability score for each abnormality
Location of each abnormality
The default cut-off is 0.6 and it can be adjustedYes, the AI results are decribed in a structured report with detail of finding name: Atelectasis, Blunted costophrenic angle, Calcification, Cavity, Consolidation, Fibrosis, Loculated pleural effusion, Mass, Nodule, Opacity, Pleural effusion, Pneumothorax
Other: Edema, Interstitial lession, Cardiomegaly, Widening mediastinum, Fracture, Medical devices, Abscess, Air fluid level, Atelectasis, Calcification, Cavity, Fibrosis, Loculated pleural effusion, Mass, Nodule, Opacity, Pleural effusion, Pneumothorax
Hardware
CPU: core i5 or above
RAM: 8 GB
Hard disk: 512 GB
Monitor: 1600 x 1200 - 21 inch
OS: Window 7 or above
Browser : Chrome
Server
CPU: Intel core i9-9980HK
RAM: 32GB
NIC Card: 2 NIC Cards - 10/100/1000Mbps
Storage: 2TB
OS: Linux Ubuntu LTS - 64bit
Internet Bandwidth (FTTH): >= 100 Mpbs
Hardware
CPU: core i5 or above
RAM: 8 GB
Hard disk: 512 GB
Monitor: 1600 x 1200 - 21 inch
OS: Window 7 or above
Browser : Chrome
Integration with X-ray Systems
Integration with PACS and Legacy Systems
DrAid™ is remotely installed at hospitals to receive and process DICOM files from X-ray imaging machines locally (on-premise) or to our VinBrain Cloud (cloud-based) via DICOM protocol (IP Address, AET title and Port)
DrAid™ can integrate seamlessly with PACS by receiving DICOM files from the PACS and sending 2nd DICOM files with notifications to the PACS.
Software
No additional software needed
Processing Time
It takes 10-15s from receiving a CXR image to returning AI result.
Data Sharing & Privacy
To maintain patient privacy, DrAid™ removes the following patient info (Patient name, Patient Address, Institution Name, +5 more) from DICOM metadata before sending to Cloud. It can't be shared to developer & deidentify.
Price
Flexible Pricing models for offline and online deployment
Pricing for DrAid is based on per-image screening.
Based on volume, there is subscription model is also available. Please contact info@vinbrain.net for more details
Public and Private system pricing is the same.
There is an initial mobilization fee in the range of $3000-$6000 based on location and mode(offline/online). Annual Maintenance Contract (AMC) and customization depending on the mode and operating environment.
Software Updates
Normally, DrAid™ do not update frequently unless we receive the customer feedback for change requests or we need to update the latest AI models. In case we update the AI models, it does not affect the user's workflow.
Product Development Method
We use state of the art methods as follows: CNN (ConvNeXT, ResNeXT), Semi-supervised Learning, Active Learning
Training
We used a total of 206k CXR images to train the product. The CXR images consist of patients above the age of 16. The demographic coverage includes datasets from SEA patients ( Vietnam, India, Indonesia, Thailand) & China
Reference Standard
We used two (2) senior radiologists with over 10 years of experience to manually annotate 110k CXR images and one (1) radiologist with 20 years of experience as an Arbitrator. The remaining annotations were collected from the TB sputum test and public dataset.
Publications
1. https://www.medrxiv.org/content/10.1101/2021.02.28.21252634v2.full.pdf
2. https://www.cureus.com/articles/79104-deep-learning-model-with-convolutional-neural-network-for-detecting-and-segmenting-hepatocellular-carcinoma-in-ct-a-preliminary-study#!/
3. An Unsupervised Learning Approach to 3D Rectal MRI Volume Registration
This website works best with browsers other than Internet Explorer.