Kassandra demonstrates how to bring machine learning models as close as possible to the clinician.
All data is processed directly in the browser, thus no information ever leaves the computer that is used to access the webpage.
Furthermore, no direct access to the imaging file is needed as the model can process the screenshot obtained from the screen-sharing window.
This project is however not all about feasibility, as it should also bring to light the potential dangers and pitfalls of machine learning models in clinical practice. Indeed, although they can add useful information, they should never substitute clinical and critical judgement.
If you wish to support this project or wish to see further models hosted in this fashion, please do not hesitate to contact me.