A comparison between Nonlinear Least Squares and Maximum Likelihood estimation for the prediction of tumor growth on experimental data of human and rat origin. (September 2019)

• Record Type: Journal Article
• Authors: Patmanidis, Spyridon; Chignola, Roberto; Charalampidis, Alexandros C.; Papavassilopoulos, George P.
• Is Part Of: Biomedical signal processing and control. Volume 54(2019)
• Publication Date: 2019-09
• Subjects: Tumor growth -- Nonlinear systems -- Parameter estimation -- Maximum Likelihood -- Nonlinear Least Squares -- Hidden Markov Model -- Noise; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101639
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Reliable growth predictions on individual basis for certain types of cancer. Consider the tumor's growth heterogeneity during the modeling procedure. More accurate predictions of rat origin tumors compared to the classical approach. Better performance for the human origin tumors compared to the classical approach. Utilization of prior knowledge to improve predictions at early growth stages. Methodology not limited to the Gompertz model. Abstract: Several mathematical models have been developed to explain the growth of tumors and used to fit experimental or clinical data. Their predictive power – i.e. their ability to forecast the future growth on the basis of present knowledge – however, has been rarely explored. Here, we investigate whether a Hidden Markov Model (HMM) based on the well-established Gompertz tumor growth function with additive Gaussian noise could effectively be used to predict the future growth of experimental tumors. The idea behind this work is that one might achieve more accurate predictions if estimates of the unknown parameters of the HMM are used instead of those obtained by fits of the deterministic Gompertz model to the data. We use the principle of Maximum Likelihood (ML) to estimate unknown parameters related to growth dynamics and noise, and we compare its effectiveness to the classical Nonlinear Least Squares minimization approach. The analyses show that our approach can provide better growth predictions when the data contain adequate information concerning the tumors saturation phase. The forecasts could also be improved by taking into account prior knowledge about the unknown parameters when the information concerning the saturation phase was inadequate. We conclude that by using HMMs in combination with the principle of ML, one can obtain more reliable growth predictions for individual tumors.