Optimization Method to Estimate Breast Tumour Parameters

Abstract

This paper presents a methodology to predict location, size and hyperactivity level of a breast tumor using temperature profile over the skin surface of the breast that may be captured by infrared thermography or numeric simulation. The estimation methodology includes an evolutionary technique based on artificial neural network (ANN), an optimization scheme based on pattern search algorithm (PSA) with linear constraints and a heat flow analysis on anatomic-accurate (realistic) breast model using finite element method (FEM). Laboratory generated datasets obtained from the FEM are applied to the ANN to associate underlying tumor with surface temperature of the model. The ANN training/testing results are in good agreement with those obtained from numeric method (FEM), thus validates the network performance. The PSA is applied for generation of solution vector sets (tumor parameters) within a given space and the solution sets are employed to produce simulated datasets using the trained ANN. The best solution set is determined by minimizing a cost function involving comparing the target temperature profiles (clinical data) to those obtained by simulation.