Last Updated:
15th May 2023
RADIFY detects tuberculosis (TB) on chest X-ray images according to radiology diagnostic standards. It highlights the pathologies related to TB with bounding boxes and heat maps, and provides a summary of the findings and categorizes images into four categories: 1) high probability, 2) intermediate probability, 3) low probability, and 4) no TB found. The images are triaged according to the highest probability of TB.
RADIFY is able to detect other pathologies on the chest X-rays that are unrelated to TB and will also highlight these pathologies with bounding boxes.
RADIFY for teleradiology is a full radiology reporting service offered as an additional service to practices with medical resource challenges.
RADIFY artificial intelligence (AI) models are capable of continual learning. When the Envisionit Deep AI shared cloud environment is used for deployment of RADIFY, continual learning is provided by default. For either offline or hybrid deployments, or installed on a dedicated cloud environment, the continual learning feature may or may not be enabled.
Demo available upon request.
Certification
FDA (pending), CE (pending), SAHPRA certified (class A)
Development Stage
On the Market
Deployment
Online, Offline, Hybrid (offline analysis, on demand synchronization with legacy systems/cloud storage)
Intended Age Group
2+ years
Target Setting
Primary health centers, general hospital (above primary level), teleradiology companies, government/public sector, e.g. national TB programme, private sector, original equipment manufacturers
Current Market
Planned market: Africa, UK, USA
Input
Chest X-ray image format: JPEG, PNG, DICOM
Chest X-ray type: posterior-anterior chest X-ray, anterior-posterior chest X-ray
Output
· Categorized output into 4 categories: 1) high probability, 2) intermediate probability, 3) low probability and 4) TB not found.
· Location of each abnormality
· Summarized AI confidence score(AIC) % on each pathology detected.
· Lung findings identified by AI: abscess, atelectasis, blunted costophrenic angle, bronchiectasis, calcification, cavity, consolidation, fibrosis, honeycombing, hyperinflation, lymphadenopathy, mass, nodule, opacity, pleural effusion, pneumothorax, cardiomegaly including cardiothoracic ratio, infiltrates, reticular infiltrates, pneumomediastinum, cysts, bone lesion, micronodules, emphysema, peribronchial thickening.
· Lung findings included in TB score: abscess, calcification, cavity, consolidation, fibrosis, lymphadenopathy, nodule, pleural effusion, prominence in the hilar region, infiltrates.
· Disease scores identified: COVID-19 pneumonia, cancer, pneumonia, cardiac disease.
· The platform can also measure the cardiothoracic ratio.
· Report and conclusion of high, intermediate, and low probability of TB.
Hardware
Offline deployment: 16 GB RAM, recommended 32 GB for picture archiving and communication system. Dual- or quad-core 64-bit (x86) CPU.
Hybrid deployment: 240 GB SSD main storage, SSD or magnetic storage for image archive. 2TB SSD main storage if continuous training is used. Nvidia Tesla-based GPU for accelerated inference.
Server
Storage: 2TB
Hardware
Offline deployment: 16 GB RAM, recommended 32 GB for picture archiving and communication system. Dual- or quad-core 64-bit (x86) CPU.
Hybrid deployment: 240 GB SSD main storage, SSD or magnetic storage for image archive. 2TB SSD main storage if continuous training is used. Nvidia Tesla-based GPU for accelerated inference.
Integration with X-ray Systems
Integration with PACS and Legacy Systems
It is possible to integrate the product with the client’s legacy picture archiving and communication system. Fee is included in the implementation fee.
Software
Should you require further information, please contact:
• info@envisionit.ai
• https://envisionit.ai/support/
Processing Time
30 milliseconds
Data Sharing & Privacy
Typically data are not shared with the manufacturer. However, if Envisionit Deep AI also provides teleradiology services as part of the engagement, data will be shared within the bounds of the agreement.
For both online and offline setups, image de-identification is performed to remove patient and/or hospital identifiers before being processed by the AI.
All received DICOM files are anonymized.
All identifiers are stored in a secure database on encrypted storage devices, regardless of whether the solution is deployed in the cloud or offline.
In case of cloud deployments, all communication to and from the cloud servers is encrypted using latest transport layer security (TLS) encryption with the use of certificates issued by a recognized Certificate Authority.
Software Updates
Every 2 months. RADIFY AI models are capable of continual learning, and, if enabled, will be trained on more frequently either in the offline/hybrid deployments or dedicated cloud environments. Envisionit Deep AI shared cloud environment provides continual learning by default.
Integration with picture archiving and communication system included in the implementation fee.
Updates are included in the license fee.
Product Development Method
Supervised deep learning (CNN, RNN)
Training
500 000 CXRs from Southern Africa, India, China, Europe, USA
Reference Standard
Human reader (radiologists specializing in TB), some are confirmed with GeneXpert
Publications
Peer-reviewed publications are not yet available.
This website works best with browsers other than Internet Explorer.