Select

Welcome to Risk-XAI: Explainable AI Models for Predicting Personalized Risk Profiles

The motivation for this demo-tool is to show, how complex AI models can be made explainable by:

  1. Disentangling most predictive features for an individual patient
  2. Mapping of features on available background knowledge

Using this demonstrator you can:

  • browse through the medical history of a patient and map diagnoses and medications on available background knowledge,
  • make individualized predictions of future risks for a set of comorbidities that are common for epilepsy patients,
  • explore most predictive factors for each individual and map them on available background knowledge about diagnoses and medications.

Personal risk profiles are estimated via DeepLORI, an AI based algorithm recently developed by the AI & Data Science team at SCAI (https://www.frontiersin.org/articles/10.3389/frai.2021.610197/full(opens in a new tab or window)). The model has been trained with data of more than 20,000 epilepsy patients observed over a time period of 1 year. It has been validated with additional data of almost 100,000 epilepsy patients.

Quick Start - Playground using simulated patients:
Real patient data is highly sensitive and cannot be made public due to legal constraints. Hence, we cannot show data or predictions for real patients here. We thus allow for exploring the potential of our algorithms via the help of simulated patients.

  1. From the top-left dropdown-menu select an arbitrary patient you like to play with. These are simulated patients, not real patients.
  2. The dashboards descriptions will give you guidance.

Quick Start – Get predictions for your own data:
Please note our Disclaimer below. You can upload your own data to this tool and receive predictions using our AI models. The required data format can be retrieved via the "Download Data" button in the top right corner in the dashboard view. The process is as follows:

  1. Upload Data: We recommend to first download the simulated data to understand the required data format.
  2. Wait 5-10 minutes for finishing the necessary calculations.*
  3. After the computations are finished, the dashboards will show up. You can start interacting with the results now.*

Note: There is absolutely NO uploaded data stored at our side, it is only buffered for allowing to perform the necessary calculations.* If you like to save the outcome, you could make screenshots.

*Refreshing the page, moving to another URL or closing the browser will delete all progress and data.