Optimal control for effective radiotherapy

Internship project at Indian Institute of Science (IISc)

This project was carried out as part of my semester long internship at the Indian Institute of Science. During this internship, I simulated an optimal controller for effective radiotherapy using Model Predictive Static Programming (MPSP). This technique aims to solve the originial nonlinear optimal control problem in a computationally efficient manner. It derives it’s ideas from Model Predictive Control (MPC) and Approximate Dynamic Programming (ADP). The initial phase of my project was to use a modified MPSP technique called Impulsive-MPSP or I-MPSP to determine the distribution of radiation dosages to drive the number of cancer cells to zero using external beam radiotherapy. To model the tumor dynamics, a two-compartment oxygen model was utilized. This was a re-implementation of a prior work in this area.

The second phase of the project involved extending this technique to account for the uncertainites in the model parameters and initial conditions. To do so, I designed a new techique known as Unscented MPSP for Impulsive Systems. This uses the technique of unscented optimal control with MPSP for impulsive systems. I formulated the mathematics for this controller to achieve the final time constraints as well as mimimize the final time variance for impulsive systems. An example prey-predator problem containing parametric uncertainties was discussed to verify the algorithm.

The above two projects are explained in greater details in the following two reports: