Created by: Nicolas Acevedo, Pablo Aldana, Marie Lenglet, Andreis Purim and Pablo Skewes
This is an older project for the M1 Elective "Inteligence Artificel et Santé" at the École Centrale de Lille. The objective was to implement something inovative at the Centre Hospitalier Universitaire de Lille (CHU Lille) combining health and IA. We decided to create a program that every time a doctor added a new prescription, it would check and detect if it is correct and/or can cause further problems for the patient.
Since the data (which can't be released publicly, sorry) we had access to were prescriptions of past patients which were CORRECT, we had to work with unsupervised learning, especially clustering. The report is in the repository. A few explicative images can also help:
Image explaining the basic idea: the doctor adds the drug prescription and the program detects incorrect ones
Some clustering with Doliprame. Since this was done with a PCA, it makes senses the clusters look slightly weird.
Word2Vec made with the Code ATC from the prescription texts
Since it is not maintained anymore (and it depends on other resources exclusive to the CHU), I'm leaving this as a legacy project.
