Influence of cost functions and optimization methods on solving the inverse problem in spatially resolved diffuse reflectance spectroscopy - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2017

Influence of cost functions and optimization methods on solving the inverse problem in spatially resolved diffuse reflectance spectroscopy

Résumé

Diffuse reflectance spectroscopy (DRS) has been acknowledged as a valuable optical biopsy tool for in vivo characterizing pathological modifications in epithelial tissues such as cancer. In spatially resolved DRS, accurate and robust estimation of the optical parameters (OP) of biological tissues is a major challenge due to the complexity of the physical models. Solving this inverse problem requires to consider 3 components: the forward model, the cost function, and the optimization algorithm. This paper presents a comparative numerical study of the performances in estimating OP depending on the choice made for each of the latter components. Mono- and bi-layer tissue models are considered. Monowavelength (scalar) absorption and scattering coefficients are estimated. As a forward model, diffusion approximation analytical solutions with and without noise are implemented. Several cost functions are evaluated possibly including normalized data terms. Two local optimization methods, Levenberg-Marquardt and TrustRegion-Reflective, are considered. Because they may be sensitive to the initial setting, a global optimization approach is proposed to improve the estimation accuracy. This algorithm is based on repeated calls to the above-mentioned local methods, with initial parameters randomly sampled. Two global optimization methods, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), are also implemented. Estimation performances are evaluated in terms of relative errors between the ground truth and the estimated values for each set of unknown OP. The combination between the number of variables to be estimated, the nature of the forward model, the cost function to be minimized and the optimization method are discussed.
Fichier non déposé

Dates et versions

hal-01588550 , version 1 (15-09-2017)

Identifiants

Citer

Prisca Rakotomanga, Charles Soussen, Walter C.P.M. Blondel. Influence of cost functions and optimization methods on solving the inverse problem in spatially resolved diffuse reflectance spectroscopy . SPIE BIOS, Jan 2017, San Francisco, CA, United States. ⟨10.1117/12.2249607⟩. ⟨hal-01588550⟩
73 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More