Watcher Documentation

Watcher is a generative AI model that simulates patient timelines. The AI takes a patient timeline as input and generates future clinical timeline.
It enables computational modeling of patient trajectories, and synthesizing large-scale patient records.
Potential applications include personalized medicine, in-silico clinical trials, and counterfactual simulations.
How the model generates patient timelines
Start with 👉 Tutorial
GitHub 👉 https://github.com/yuakagi/Watcher

Model Overview

Watcher model architecture
It encodes categorical, numeric, and temporal entries into vectors, which are processed by decoder-only transformer layers.
The model autoregressively generates future clinical events, one event at a time.
At each step, the model outputs a probability distribution over its entire vocabulary.

Digital-Twin EHR System

This is our proof-of-concept digital-twin product, and our vision for the future of AI in healthcare.
Watcher serves as the backend simulator in the system (Figure):
Watcher model architecture
This system enables the simulation and real-time feedback through an interactive web application.
It consists of three components:

  1. AI model [GitHub]: A generative model that simulates patient timelines. (👈 This package you are currently reading.)

  2. Digital-twin EHR [GitHub]: A web-based, AI-powered EHR that interacts with the model and visualizes simulation results.

  3. Data pipeline: A data pipeline that supplies real-world clinical data to the data server.

To try the full digital-twin system, please follow these steps:

Step 1: Prepare your clinical data

  • Required clinical data are defined in Clinical Records

  • You can use your own clinical data or publicly available datasets.

Step 2: Upload clinical data to database

  • Watcher package provides a docker container for PostgreSQL database.

  • You can upload your clinical data to the database using the package.

Step 3: Train the AI model

  • Train (pretrain & fine-tune) the AI model using the Watcher package following the tutorial.

Step 4: Launch the simulation API server

  • The Watcher package provides a simulation API server that runs the AI model (gunicorn + Flask).

  • This will be the API server that the digital-twin EHR system will communicate with.

  • Launch the server following the tutorial.

Step 5: Launch the digital-twin EHR system

  • TwinEHR is a web application that provides a user interface for the simulation API.

  • Clone the repository and set proper environment variables to connect to the simulation API server.

  • Run the web application server

Note

  • For Japanese users, our data pipeline is available to conveniently collect and clean clinical data, but its use is not mandatory.

  • Users outside Japan can also use the system with your own clinical data or publicly available datasets.

Documentation