SCRO / Journées de l'optimisation

HEC Montréal, 29-31 mai 2023


HEC Montréal, 29 — 31 mai 2023

Horaire Auteurs Mon horaire

AORT Applied Optimization for Radiation Therapy

30 mai 2023 10h30 – 12h10

Salle: Hélène-Desmarais (bleu)

Présidée par Danielle Ripsman

4 présentations

  • 10h30 - 10h55

    A Time-Discretized Mathematical Model and Heuristic for Breast Cancer Radiotherapy

    • Scholar Sun, prés., University of Waterloo
    • Ernest Osei, Grand River Regional Cancer Centre
    • Johnson Darko, Grand River Regional Cancer Centre
    • Houra Mahmoudzadeh, University of Waterloo

    Intensity modulated radiation therapy (IMRT) is a cancer treatment method whereby a high-energy beam is emitted by a linear accelerator (LINAC) to irradiate cancerous tissue. Sliding window IMRT is a technique in which the beam is modulated by a set of tungsten leaves that move unidirectionally across the beam field to produce complex delivery patterns. Planning for this treatment involves specifying a sequence of leaf arrangements and the corresponding time the LINAC takes on this arrangement. Due to the exponentially large number of possible arrangements, this type of treatment can be difficult to plan.

    In this talk, we formulate a mixed integer nonlinear program to optimize the delivery of a sliding-window IMRT procedure wherein the prescribed dosage to the tumours are met and the dose to healthy tissues are minimized. Given the complex nature of the nonlinear MIP model, we propose a heuristic that provides a high-quality feasible solution which can be used as a benchmark for clinical plans. The viability of this model is then demonstrated using patient data. The solutions generated by the model can be directly implemented in clinical treatment planning software to assist planners in improving the quality of plans for cancer patients.

  • 10h55 - 11h20

    Generating Gamma Knife Treatment Plans Using Knowledge-Based Planning with Inverse Optimization

    • Binghao Zhang, prés., University of Toronto
    • Aaron Babier, University of Toronto
    • Mark Ruschin, Sunnybrook Health Sciences Centre
    • Timothy Chan, University of Toronto

    Current methods for GK treatment planning utilizes either manual forward planning, where planners manually place shots in a tumor to achieve a desired dose distribution, or inverse planning, whereby the dose delivered to a tumor is optimized for multiple objectives based on established metrics. For other treatment modalities like IMRT and VMAT, there has been a recent push to develop knowledge-based planning (KBP) pipelines to address the limitations presented by forward and inverse planning. However, for GK, there has been comparatively little work done in creating a full KBP pipeline.

    In this talk, we present a novel GK-specific KBP pipeline utilizing inverse optimization in conjunction with 3-dimensional dose prediction for the generation of deliverable GK treatment plans. This is the first KBP pipeline that applies a generalized inverse optimization formulation to optimize GK quality metrics. We compare its efficacy to a dose mimicking model as well as to plans resulting from manual forward planning. Additionally, we characterize the performance of the optimization models with respect to the associated predictions. We show that this pipeline allows for plans to consistently achieve superior levels of quality without human intervention during the plan optimization stage.

  • 11h20 - 11h45

    Superior and Light Pareto Robust Optimization in Radiation Therapy

    • Fahimeh Rahimi, prés., University of Waterloo
    • Hossein Abouee Mehrizi, University of Waterloo
    • Houra Mahmoudzadeh, University of Waterloo

    Robust optimization (RO) is a well-known methodology used in various applications where uncertainties exist. However, the over-conservative nature of the RO approach may not always be desirable, especially when non-worst-case scenarios are important. In contrast, Pareto robust optimization (PRO) offers a way to balance worst-case and non-worst-case performance, generating robust solutions that are less conservative for non-worst-case scenarios.
    In this talk, we present an extension of PRO that leverages expert knowledge on likely uncertain scenarios to improve its applicability in practical settings. Specifically, we define and characterize superior PRO solutions and develop algorithms to identify the Pareto robust frontier of these solutions. We also introduce the concept of light PRO solutions that trade off the worst-case performance to improve non-worst-case behaviour while maintaining robustness and quantify the trade-off between worst-case optimality and non-worst-case performance. Finally, we illustrate the application of the proposed approach to radiation therapy treatment planning for breast cancer using realistic patient datasets. The results show that the proposed approach can significantly improve the quality of the solution in non-worst-case scenarios while maintaining robustness and limiting the optimality error of the worst-case outcome within a negligible threshold.

  • 11h45 - 12h10

    A Geometric Approach to Beam Angle Optimization in Radiation Therapy Treatment Planning

    • Danielle Ripsman, prés., University of Waterloo
    • Sibel Alumur Alev, University of Waterloo
    • Houra Mahmoudzadeh, University of Waterloo

    Beam angle optimization (BAO) is a difficult but essential component of planning many types of radiation therapy treatment. Despite the wealth of proposed methodologies for optimal beam-angle selection in the literature with significant treatment quality gains, in practice, clinicians often opt for the selection of a fixed number of equidistant beams, or manual iterative planning. This is due, in part, to the requirement for a secondary fluence map optimization (FMO) to validate any BAO selections and the resource-intense calculations needed to calculate the parameters for such a model at each iteration.
    In this talk, the BAO problem is modeled using a geometrical abstraction, allowing it to be considered in a single-stage column generation-driven set-covering framework. This novel abstraction allows for a reduction in the reliance of BAO modeling on sophisticated dose calculators, as well as eliminating the need for time-consuming BAO-FMO iteration.