Periodic Reporting for period 1 - CoronAI (AI software for the automated detection of covid-19 infection on chest CT)
Okres sprawozdawczy: 2020-04-01 do 2021-06-30
Aidence is creating a new product to automatically analyse chest CT scans and diagnose the presence of corona (covid-19) viral infection. The product receives an image from the CT scanner, analyses it, and returns an easy-to-read report on the most likely diagnosis. This is done in the background fully automatically and the report is available for medical staff within a few minutes.
Studies indicate CT is as least as sensitive as the current RT-PCR kit for testing and is very fast. It is useful in case no kits are available, or when they are inconclusive, and/or radiology staff is unavailable or untrained.
This product can be used for triaging as well as longer-term monitoring and early warning for future outbreaks. Because the solution will be operating on all CT chest scans and fully automated, once deployed in hospitals around Europe it will also serve as a case registry for management reporting and tracking of a new potential corona outbreak.
Studies indicate CT is as least as sensitive as the current RT-PCR kit for testing and is very fast. It is useful in case no kits are available, or when they are inconclusive, and/or radiology staff is unavailable or untrained.
This product can be used for triaging as well as longer-term monitoring and early warning for future outbreaks. Because the solution will be operating on all CT chest scans and fully automated, once deployed in hospitals around Europe it will also serve as a case registry for management reporting and tracking of a new potential corona outbreak.
Since the start of the project, we have built two AI models to detect Covid on Chest CT scans and another model to determine the severity of Covid in a patient. We have rebuilt our existing single AI product engine (capable of detecting lung cancer nodules on Chest CT’s) to a multiple AI product engine, ready to link the Covid AI models to the new engine.
As the pandemic was creating all kind of uncertainties for the health system, we decided to develop our AI models in parallel with rebuilding the engine. This reduced the development horizon from 2-3 years down to 15 months. We collected the scans of Covid patients in a joint effort with hospitals all over Europe and named the project ICOVAI. It took us longer than estimated to collect the scans and we had to build an annotator tool in order to be able to correctly annotate the scans. However, the model, was ready for clinical validation testing by the end of 2020 (9 months development time). We contracted NKI/AVL, a very respected Dutch Health Institute to perform the clinical validation study. This was started in Q1 2021. Although NKI/AVL was extremely positive on the way our severity model could indicate the severity of Covid in patients, the clinical validation study also revealed that the AI model did not distinguish enough with other lung (infection) diseases.
It was our ambition that by the time the model was clinically validated we would install the AI software at 50 sites. All existing sites will be transferred to the new engine by Q2 2021. By the end of June 2021, we have over 50 sites contracted and ready to receive the AI Covid model on the new engine. However, with the negative outcome of our clinical validation study we decided not to launch the product. At 30th of June we informed our customers and all other hospitals that supplied us with scans on the outcome of our project.
As the pandemic was creating all kind of uncertainties for the health system, we decided to develop our AI models in parallel with rebuilding the engine. This reduced the development horizon from 2-3 years down to 15 months. We collected the scans of Covid patients in a joint effort with hospitals all over Europe and named the project ICOVAI. It took us longer than estimated to collect the scans and we had to build an annotator tool in order to be able to correctly annotate the scans. However, the model, was ready for clinical validation testing by the end of 2020 (9 months development time). We contracted NKI/AVL, a very respected Dutch Health Institute to perform the clinical validation study. This was started in Q1 2021. Although NKI/AVL was extremely positive on the way our severity model could indicate the severity of Covid in patients, the clinical validation study also revealed that the AI model did not distinguish enough with other lung (infection) diseases.
It was our ambition that by the time the model was clinically validated we would install the AI software at 50 sites. All existing sites will be transferred to the new engine by Q2 2021. By the end of June 2021, we have over 50 sites contracted and ready to receive the AI Covid model on the new engine. However, with the negative outcome of our clinical validation study we decided not to launch the product. At 30th of June we informed our customers and all other hospitals that supplied us with scans on the outcome of our project.
The impact for Aidence has been very strong. First of all, it has strengthened our reputation as the AI specialist on Chest CT’s. The Aidence model was developed according to the latest insights regarding the proper clinical validation of medical AI (confirmed by AVL/NKI). So, from a scientific perspective this strengthens our reputation. This is very valuable as we are a 6 year old start up/scale-up.
The publication of the scientific results contributes to the knowledge at Aidence, NKI and the wider scientific community regarding the best practices and pitfalls of medical AI development, impacting subsequent product quality and thereby improving patient outcomes.
The annotated dataset is anticipated to have additional impact on the accurate detection of lung cancer, thanks to the overlap between AI models needed for Covid and lung cancer. This impacts the earlier detection of lung cancer and thereby improving patient survival rates.
From a commercial perspective the project has helped us building a customer base and achieving commercial traction. With this traction we are planning to get a B Round and to further invest in developing AI solutions that will help transforming our healthcare system.
The publication of the scientific results contributes to the knowledge at Aidence, NKI and the wider scientific community regarding the best practices and pitfalls of medical AI development, impacting subsequent product quality and thereby improving patient outcomes.
The annotated dataset is anticipated to have additional impact on the accurate detection of lung cancer, thanks to the overlap between AI models needed for Covid and lung cancer. This impacts the earlier detection of lung cancer and thereby improving patient survival rates.
From a commercial perspective the project has helped us building a customer base and achieving commercial traction. With this traction we are planning to get a B Round and to further invest in developing AI solutions that will help transforming our healthcare system.