Text by:ITCL TECHNOLOGY CENTRE
The Benchmarking framework focuses on providing valid tools to the different pilots for the analysis of their data from an initial point before deploying the systems and the subsequent deployment and commissioning of the pilots and the proportion of the ranking values for the Marketplace.
The Benchmarking system is open to the collection of data from other prototypes outside the HosmartAI project, by sending the same inputs as the current pilots. With this, the developed models for measuring performance of AI solutions can be used, and the corresponding rankings will be calculated and published in the Marketplace. For example, if one of the developed models predicts the medical cost of a new patient from the beginning of their treatment, other hospital managers outside the HosmartAI project, could use this model to determine the cost of their own patients, considering the variables used initially (number of sessions, severity, etc…).
The selection of the algorithms that allow the different pilots for data analysis and improvement of the system has been started. For this, in a first selection, general purpose algorithms were selected allowing the analysis of the existing correlations between the KPIs defined by the pilots and the variables collected by sensors and medical information record systems (calendars, treatments, diagnoses). Thus, through these algorithms, it is intended that the pilots can identify the variables that most influence their results and their objectives in order to improve their developments throughout the entire HosmartAI project.
The Benchmarking framework will allow users of the different pilots to process the data on a Jupyter platform and generate models, validate them, and visualize the results. Subsequently, these models can be integrated into the Benchmarking framework and allow their use in other developments that have the same characteristics.
The use of the framework will be merged into the HosmartAI partners’ work as from the beginning, and new participants are enabled to compare the HosmartAI pilots with their own pilots.