- Daniel Kompe
- Anne Schwerk
- sebastian friedrich
- Christoph Hanser
- Xuanpu Sun
- Nadia Witte
- Theresa Freund
- franky Sommer
- Stefan Krojer
- frank sommer
- Arunakiry Natarajan
- Matthias Buchhorn-Roth
- Domitila Barros
- Sami Hassan
Social added value
The target group of our Open Source solution are people with direct contact to COVID-19 patients*, e.g: Doctors and catering staff. It solves the urgent need for automated quality control of imported respirators, so that they can be purchased in large quantities even from unknown manufacturers from abroad and if necessary without CE marking. Currently, clinics can only use FFP2 masks and only to a limited extent, so that unwanted exposure has become essential and many have already been infected with COVID-19.
Our team started in the COVID19 Hackathon at the world-wide largest hackathon of the German Government #WirvsVirus on week ago.
The project will search counterfeit FFP-2/FFP-3 masks in the fraud databases of the EU, UN, USA, China and independent national testing bodies such as TÜV in order to achieve better accuracy by integrating the results. This meta search is to be used for the detection of subsequent offenders as well as for the extraction of counterfeit features. This data will be combined with further background data of the suppliers (from websites and CE-certificates) to generate a risk score that assesses the credibility of the manufacturers. This initial check of the background data of the suppliers is not feasible under the current crisis situation and lack of personnel and would be time consuming even for experienced personnel and would exceed their expertise.
In a second step, the masks already ordered can be checked for authenticity on site with CustomVisionAI. Currently, a visual inspection of the masks is carried out manually by the hygiene departments of hospitals, with little standardization and limited time resources. In general, there are standardized procedures for checking the masks for their condition and whether the masks meet the requirements. Currently, however, hospitals do not receive samples prior to delivery, so that testing can only take place once the masks have been delivered. Since delivery times currently take up to 3 weeks and final approval would then take up to 4 weeks, masks with insufficient verification must be used.
At the moment, the demand for masks is more than unmet: hospitals are running out of FFP2 masks, so they can only be used for the most urgent cases. The FFP3 masks required for COVID-19 are no longer in stock. Forecasts show that the COVID-19 pandemic could last for months or years and affect the entire world population. Even after the crisis, the solution will be able to be used in routine operations and shows a potential transfer option to other medical products, such as protective suits, goggles and disinfectants. Other countries can also benefit from the open source solution.
Progress during the Hackathon
The interdisciplinary team of experts* for hospitals, medical equipment, digital health, purchasing processes, logistics, AI and IT have carried out in-depth research, compiled and discussed in recent days.
In the hackathon, the existing black lists of national, European and international fraud cases were searched for FFP-2/FFP-3 masks and important characteristics of the fraud cases were extracted and collected. The fraud databases of the EU, USA, China as well as independent testing agencies like TÜV Süd have shown that different sources have to be used to detect fraud and to make maximum use of the existing knowledge.
The following databases are searched and extracted:
Other contact points for fraud prevention
Images of real and fake FFP2/FFP3 masks were collected and uploaded to Azure CustomVisionAI. This was used to train the first model and then tested with a fixed set of test data.
With the prototype, the user* can upload images via his/her smartphone and immediately receive the result from the cloud-based AI service (Azure Custom Vision AI).
The model can be tested on the Test page of the prototype.
In the open source GitHub repository, the source code can be freely obtained and developed: https://github.com/chaosmail/is-my-mask-safe
Comprehensibility of the solution
The current crisis situation requires that all opportunities to purchase respiratory masks be used. As a result, clinics and hospitals are increasingly purchasing imported Chinese respirators, which have CE certificates, but which often cannot be verified due to scarcity of resources and complexity. Many cases of fraud have already been uncovered and stored in databases of TÜV Süd, among others.
The solution should be easy and intuitive to use by hospital staff. In the first step, the purchasing department can check the website and associated information (server locations, missing imprint, incorrect telephone/address details, website traffic, time of website presence, pricing, etc.) by entering the URL of the provider. The solution also pulls the information from the website and compares it with fraud databases.
In a next step, additional information of the provided CE certificates is checked and provided with a further risk score. This includes the verification of the international tax ID (VAT) and its validity on the VIES website of the European Commission by means of: Vatlayer]( https://vatlayer.com) or VAT-validationVAT-validation as well as the verification of the authenticity of the CE symbol and the corresponding number of the notified inspection bodies. If these initial checks have resulted in a positive risk score, masks can be ordered, which are then visually checked for quality by AI in a third step on arrival. The images are captured by the smartphone, stored for documentation purposes and pass through a multi-stage automatic quality process using various AI-based data models. Here, classification as well as object detection is important. Subsequently, the user(s)* are again informed about certain quality criteria of the specific model, which cannot be achieved by image recognition, e.g. number of layers and smell.
The research results were documented in detail in Microsoft Teams and OneNote. The model was stored in the cloud and further project participants can be invited.
In the current crisis situation, hospitals are forced to use masks from non-verified importers and thus expose themselves to potential COVID-19 infections. According to the European Commission EU Commission recommendation, the CE seal is no longer mandatory for the use of respiratory masks. Thus, the verification of manufacturers is of enormous relevance and should be automatically checked.
What impact for people
The solution protects especially vulnerable professionals and ensures safe and sustainable work with potentially infected patients* and thus the continuity of our medical care.
Try It out
azure, azurecognitiveservices, microsoft, powerapps, powerplatform, procurement, supplier, teams