Multiobjective decomposition of integer matrices: application to radiotherapy

TL;DR

This paper addresses a complex multiobjective problem in radiotherapy, decomposing integer matrices into binary matrices with the consecutive ones property to optimize treatment delivery, using a heuristic approach.

Contribution

It introduces a heuristic based on Pareto local search for decomposing integer matrices in radiotherapy, tackling an NP-hard multiobjective problem.

Findings

Heuristic effectively decomposes matrices for various sizes.

Results show improved trade-offs among objectives.

Experiments validate the approach's practicality.

Abstract

We consider the following problem: to decompose a nonnegative integer matrix into a linear combination of binary matrices that respect the consecutive ones prop- erty. This problem occurs in the radiotherapy treatment of cancer. The nonnegative integer matrix corresponds to fields giving the different radiation beams that a linear accelerator has to send throughout the body of a patient. Due to the in- homogeneous dose levels, leaves from a multi-leaf collimator are used between the accelerator and the body of the patient to block the radiations. The leaves positions can be represented by segments, that are binary matrices with the consecutive ones property. The aim is to find efficient decompositions that simultaneously minimize the irradiation time, the cardinality of the decomposition and the setup-time to configure the multi-leaf collimator at each step of the decomposition. We propose for this NP-hard multiobjective combinatorial problem a heuristic, based on the adaptation of the two-phase Pareto local search. Experiments are carried out on different size instances and the results are reported.