A Cloud Application for monitoring AKI in ICU patients

U-Care Medical is a spin-off of the Polytechnic of Turin that operates in the biomedical field and whose core business is to transform kidney care through AI technology. The main idea is to use Machine Learning models to predict the probability of ICU (Intensive Care Unit) for patients contracting a severe, and in some cases persistent, form of AKI (Acute Kidney Injury). This thesis, made in collaboration with the above spin-off, aims at creating an easy-to-use Proof of Concept targeting the ICU medical staff, which should be able to access the platform, view ICU patients’ and possibly inspect them individually. In this case, their vital parameters such as diuresis, creatinine, bicarbonate, and more are shown, using graphs that show their hourly distribution. The application was designed to be in the cloud. In this regard, it is structured following a microservices architecture. The main workflow starts after medical staff, such as doctors or nurses, log into the application. At this point they can see the main page exposed by the frontend service with all the ICU patients and inspect the status of each of them. The final platform should interact directly with the EHR system (Electronic Health Record) of the ICU, in this way new data is injected into the system. The backend is in turn structured in small independent services that perform different functions. The Endpoint Service and GraphQL Service expose REST and GraphQL APIs respectively. The other services, Preprocessing, Staging, Akira and Persea deal with interacting with the Data Processing service which invokes the appropriate Python scripts. Since these data are supposed to be raw, they should be preprocessed thanks to the Data Processing service. It is essential that AKI is detected early and treated promptly due to the problems that may arise with this disease. Indeed, without quick treatment, the normal function of other organs could be compromised. This application will powerfully help the ICU medical staff to recognize in advance whether a patient will contract or not a form of AKI, thus helping to start therapies early. In a future perspective, the application could be easily scaled in order to handle multiple hospitals. In this case, there must be no leakage of data between patients belonging to different hospitals. Furthermore, the application could be extended to introduce new functionalities, such as the support for other diseases whose timely intervention may be essential, and the possibility to handle multiple hospitals at the same time, of course without having data breaches between the different hospital facilities.