SCRO / Journées de l'optimisation

HEC Montréal, 29-31 mai 2023


HEC Montréal, 29 — 31 mai 2023

Horaire Auteurs Mon horaire

ILH Inverse Learning in Healthcare

30 mai 2023 15h30 – 17h10

Salle: Hélène-Desmarais (bleu)

Présidée par Kimia Ghobadi

3 présentations

  • 15h30 - 15h55

    A tractable approach to inverse optimization under Euclidean norm

    • Sara Ebrahimkhani, prés.,
    • Hossein Abouee Mehrizi, University of Waterloo
    • Houra Mahmoudzadeh, University of Waterloo

    In this talk, we propose an inverse optimization method to infer the feasible region of a linear problem based on multiple observations as input. We introduce a measure to evaluate the goodness-of-fit of the inferred feasible region based on its Euclidean distance from the observations. Given the nonlinear nature of the Euclidean distance metric, we propose a non-smooth exact penalty approach to find the exact optimal solution of the problem tractably. We demonstrate an example application using laboratory report data from hospitalized patients to infer the guidelines based on which the patients are discharged.

  • 15h55 - 16h20

    Inverse Optimization for Inferring Clinical Criteria for Radiation Plans for Prostate Cancer Patients

    • BRADLEY HALLETT, prés.,
    • Ernest Osei, Grand River Hospital
    • Johnson Darko, Grand River Hospital
    • Houra Mahmoudzadeh, University of Waterloo

    Our research proposes using an inverse optimization technique to understand the implicit logic of the oncologist's approved treatment plans. We collected radiation plans from a set of prostate cancer patients. We retrieved various features of successful historical plans from critical organs as well as the planning target volume (PTV) and critical target volume (CTV). Through inverse optimization, we can leverage past radiation plans to determine the updated clinical guidelines and any other unrecognized trade-offs utilized.

  • 16h20 - 16h45

    Inverse learning of diet recommendations

    • Farzin Ahmadi, Johns Hopkins Universiry
    • Fardin Ganjkhanloo, Johns Hopkins Universiry
    • Kimia Ghobadi, prés., Johns Hopkins University

    We use 'Inverse learning', an inverse optimization approach, to find personalized diets for patients who suffer from hypertension. This data-driven approach considers prior food intakes of patients and embeds that within dietary frameworks that are recommended by clinical providers. The results are diets that mimic patients' habits and behaviour while gradually moving them towards healthier diets.