Updated: Sep 3
COVID-19 and TB are two of the biggest challenges in healthcare the world is facing. While COVID-19 is a recent onset pandemic, TB has been around for 15000 years and is still affecting over 10 million people annually.
Large-scale and speedy screening for such widespread diseases is extremely crucial not only for isolating patients to stop the spread of the infection but also to provide timely treatment. This large-scale screening, where all individuals in a target population group (a group considered to be at high risk) are tested in an organised manner, is called population screening. For population screening, tests like RT-PCR (for COVID-19) and GeneXpert (for TB) are reliable, but also expensive, time-consuming, and complicated to carry out. These tests are overburdening healthcare systems around the world, and ancillary diagnostic solutions can help reduce the burden.
The chest X-ray is not just a diagnostic tool, but it can also assess response to treatment, evaluate disease progression, and predict outcomes. This entire radiology workflow can be automated using deep learning algorithms.
A chest X-ray with signs of TB (red box), detected by our deep learning model.
Genki is our artificial intelligence-assisted population screening platform. Genki focuses on population screening with an aim to eliminate large-scale diseases for a healthier society. In Japanese, Genki means healthy and full of life. We use AI to improve the quality of diagnosis, reduce turnaround time, and prevent radiologist burnout, especially when dealing with large volumes of medical imaging scans in population screening programs.
Genki is being used extensively in COVID-19, tuberculosis, and other public screening programs at hospitals, in remote locations, and hotspots.
Genki evaluates X-rays and CT scans and automatically checks for potential infections like COVID-19 and TB. It triages patients in a matter of seconds, thus reducing the need for additional clinical assessment tests and in turn simplifying caregivers’ workflows. The AI delivers better accuracies while substantially reducing the cost of screening and improving response time.
We have made the case for the use of radiography and AI to screen infections. However, the wide deployment of these models in public health space comes with some unique challenges.
In population screening programs, images are collected from multiple sources and small mobile machines. This often leads to poor quality data and a common machine learning application problem called domain shift. Domain shift occurs when the distribution of images used for training a model does not match the distribution the model encounters when deployed. Thus, imaging characteristics vary from one location to another due to differences in equipment, imaging settings, patient position, etc. Even at one imaging centre, they may vary over time. The images below demonstrate how chest X-rays taken at different hospitals can look very different.
Images from three different datasets, top to bottom: CHAI, Padchest, and one of our partner hospitals, with their corresponding pixel intensity histograms.
Additionally, chest X-rays often capture foreign objects like chest leads and pacemakers, which can throw off AI models trained to detect specific pathologies such as granuloma and cause them to give a higher number of false positives.
Chest X-rays with Various Foreign Objects
We took the time to address these deployment problems. Our models include a domain shift detector and a foreign object detector. We also used a deep learning tool called “attention” to increase our models’ sensitivity. This technique makes the algorithm pay more attention to the important parts of a chest X-ray image. The output the model gives is thus more refined.
Our radiography and AI solution is painstakingly designed to manage epidemic and pandemic situations successfully. The solution can be deployed on mobile equipment, so patients in hotspots can be screened at the doorstep. In the case of TB, the provision of care gap between patients diagnosed with lab-confirmed TB and those who fail to start treatment is quite significant. Using deep learning and edge-assisted solutions, we can prevent pre-diagnostic loss to follow-up and default of patients and flag them to undergo further testing. Healthcare providers can also follow-up with patients on the phone through a call centre. The platform’s smart notifications also remind patients of follow-up clinical exams after a predefined time interval.
The Genki Solution Architecture
After instant triaging, radiologists and imaging experts can step in to evaluate the automated reports generated in compliance with WHO/RSNA/ACR guidelines by the platform and edit and submit them. This not only reduces radiologist effort significantly but also fortifies the AI model. We call this self-learning approach expert-in-the-loop, wherein Genki’s models constantly absorb radiologist feedback and become more robust. The platform automatically sends approved reports to hospitals, keeping the reporting process hassle-free.
Genki connects with government databases for disease monitoring (like Nikshay in India) to facilitate web-based patient surveillance, tracking, and monitoring in high risk areas; this is the crux of large-scale disease management.
Together, these features support treatment planning and disease containment, which are crucial for governments and NGOs working to resolve public health crises. More information on Genki features like smart reporting, smart analytics, and smart notifications can be found here - https://www.deeptek.ai/post/augmento-a-cloud-and-ai-powered-radiology-optimization-platform
The Genki Deep Learning Solution has demonstrated its usefulness as a last-mile approach in disease diagnostics. It has the potential to radically change the way TB screening and surveillance will be carried ahead. We worked with the Greater Chennai Corporation on their TB Free Chennai initiative, where we deployed on mobile vans for active surveillance. The success of this screening program has been phenomenal. We evaluated 85000 scans, and our system successfully captured 90% of TB-positive cases, proving our model’s high sensitivity.
Images from mobile vans used to screen TB, from the TB Free Chennai Project
Genki also delivered a cost-benefit by limiting the number of cases to be forwarded for further tests, like GeneXpert. This can substantially reduce costs. Genki also has a dual purpose considering the platform can tackle both COVID-19 and TB simultaneously.
AI pre-screening and triage is just one aspect of the platform. It has been tested extensively for tuberculosis, pneumonia, and other chest ailments. We believe that the validations can pave the way for using the AI-infused platform for handling large datasets in busy hospital and ICU settings on one side and also support village-to-village chest X-ray screening in mobile van screening for TB diagnosis.
We are, essentially, augmenting radiology, democratising radiology.