The optimization of microbial functions through rational environmental manipulations.

Microorganisms play a central role in biotechnology and it is key that we develop strategies to engineer and optimize their functionality. To this end, most efforts have focused on introducing genetic manipulations in microorganisms which are then grown either in monoculture or in mixed-species consortia. An alternative strategy to optimize microbial processes is to rationally engineer the environment in which microbes grow. The microbial environment is multidimensional, including factors such as temperature, pH, salinity, nutrient composition, etc. These environmental factors all influence the growth and phenotypes of microorganisms and they generally "interact" with one another, combining their effects in complex, non-additive ways. In this piece, we overview the origins and consequences of these "interactions" between environmental factors and discuss how they have been built into statistical, bottom-up predictive models of microbial function to identify optimal environmental conditions for monocultures and microbial consortia. We also overview alternative "top-down" approaches, such as genetic algorithms, to finding optimal combinations of environmental factors. By providing a brief summary of the state of this field, we hope to stimulate further work on the rational manipulation and optimization of the microbial environment.

Optimizing Antenna Positioning for Enhanced Wireless Coverage: A Genetic Algorithm Approach.

The precise placement of antennas is essential to ensure effective coverage, service quality, and network capacity in wireless communications, particularly given the exponential growth of mobile connectivity. The antenna positioning problem (APP) has evolved from theoretical approaches to practical solutions employing advanced algorithms, such as evolutionary algorithms. This study focuses on developing innovative web tools harnessing genetic algorithms to optimize antenna positioning, starting from propagation loss calculations. To achieve this, seven empirical models were reviewed and integrated into an antenna positioning web tool. Results demonstrate that, with minimal configuration and careful model selection, a detailed analysis of antenna positioning in any area is feasible. The tool was developed using Java 17 and TypeScript 5.1.6, utilizing the JMetal framework to apply genetic algorithms, and features a React-based web interface facilitating application integration. For future research, consideration is given to implementing a server capable of analyzing the environment based on specific area selection, thereby enhancing the precision and objectivity of antenna positioning analysis.

Hyperparameter Optimization with Genetic Algorithms and XGBoost: A Step Forward in Smart Grid Fraud Detection.

This study provides a comprehensive analysis of the combination of Genetic Algorithms (GA) and XGBoost, a well-known machine-learning model. The primary emphasis lies in hyperparameter optimization for fraud detection in smart grid applications. The empirical findings demonstrate a noteworthy enhancement in the model's performance metrics following optimization, particularly emphasizing a substantial increase in accuracy from 0.82 to 0.978. The precision, recall, and AUROC metrics demonstrate a clear improvement, indicating the effectiveness of optimizing the XGBoost model for fraud detection. The findings from our study significantly contribute to the expanding field of smart grid fraud detection. These results emphasize the potential uses of advanced metaheuristic algorithms to optimize complex machine-learning models. This work showcases significant progress in enhancing the accuracy and efficiency of fraud detection systems in smart grids.

An Influence of Actuator Gluing on Elastic Wave Excited in the Structure.

In this article, the practical issues connected with guided wave measurement are studied: (1) the influence of gluing of PZT plate actuators (NAC2013) on generated elastic wave propagation, (2) the repeatability of PZT transducers attachment, and (3) the assessment of the possibility of comparing the results of Laser Doppler Vibrometry (LDV) measurement performed on different 2D samples. The consideration of these questions is crucial in the context of the assessment of the possibility of the application of the guided wave phenomenon to structural health-monitoring systems, e.g., in civil engineering. In the examination, laboratory tests on the web of steel I-section specimens were conducted. The size and shape of the specimens were developed in such a way that they were similar to the elements typically used in civil engineering structures. It was proved that the highest amplitude of the generated wave was obtained when the exciters were glued using wax. The repeatability and durability of this connection type were the weakest. Due to this reason, it was not suitable for practical use outside the laboratory. The permanent glue application gave a stable connection between the exciter and the specimen, but the generated signal had the lowest amplitude. In the paper, the new procedure dedicated to objective analysis and comparison of the elastic waves propagating on the surface of different specimens was proposed. In this procedure, the genetic algorithms help with the determination of a new coordinate system, in which the assessment of the quality of wave propagation in different directions is possible.

Metaheuristic-Based Feature Selection Methods for Diagnosing Sarcopenia with Machine Learning Algorithms.

This study explores the efficacy of metaheuristic-based feature selection in improving machine learning performance for diagnosing sarcopenia. Extraction and utilization of features significantly impacting diagnosis efficacy emerge as a critical facet when applying machine learning for sarcopenia diagnosis. Using data from the 8th Korean Longitudinal Study on Aging (KLoSA), this study examines harmony search (HS) and the genetic algorithm (GA) for feature selection. Evaluation of the resulting feature set involves a decision tree, a random forest, a support vector machine, and naïve bayes algorithms. As a result, the HS-derived feature set trained with a support vector machine yielded an accuracy of 0.785 and a weighted F1 score of 0.782, which outperformed traditional methods. These findings underscore the competitive edge of metaheuristic-based selection, demonstrating its potential in advancing sarcopenia diagnosis. This study advocates for further exploration of metaheuristic-based feature selection's pivotal role in future sarcopenia research.

Customizable Colorimetric Sensor Array via a High-Throughput Robot for Mitigation of Humidity Interference in Gas Sensing.

One challenge for gas sensors is humidity interference, as dynamic humidity conditions can cause unpredictable fluctuations in the response signal to analytes, increasing quantitative detection errors. Here, we introduce a concept: Select humidity sensors from a pool to compensate for the humidity signal for each gas sensor. In contrast to traditional methods that extremely suppress the humidity response, the sensor pool allows for more accurate gas quantification across a broader range of application scenarios by supplying customized, high-dimensional humidity response data as extrinsic compensation. As a proof-of-concept, mitigation of humidity interference in colorimetric gas quantification was achieved in three steps. First, across a ten-dimensional variable space, an algorithm-driven high-throughput experimental robot discovered multiple local optimum regions where colorimetric humidity sensing formulations exhibited high evaluations on sensitivity, reversibility, response time, and color change extent for 10-90% relative humidity (RH) in room temperature (25 °C). Second, from the local optimum regions, 91 sensing formulations with diverse variables were selected to construct a parent colorimetric humidity sensor array as the sensor pool for humidity signal compensation. Third, the quasi-optimal sensor subarrays were identified as customized humidity signal compensation solutions for different gas sensing scenarios across an approximately full dynamic range of humidity (10-90% RH) using an ingenious combination optimization strategy, and two accurate quantitative detections were attained: one with a mean absolute percentage error (MAPE) reduction from 4.4 to 0.75% and the other from 5.48 to 1.37%. Moreover, the parent sensor array's excellent humidity selectivity was validated against 10 gases. This work demonstrates the feasibility and superiority of robot-assisted construction of a customizable parent colorimetric sensor array to mitigate humidity interference in gas quantification.

Parameter grouping and co-estimation in physiologically based kinetic models using genetic algorithms.

Physiologically based kinetic (PBK) models are widely used in pharmacology and toxicology for predicting the internal disposition of substances upon exposure, voluntarily or not. Due to their complexity, a large number of model parameters need to be estimated, either through in silico tools, in vitro experiments, or by fitting the model to in vivo data. In the latter case, fitting complex structural models on in vivo data can result in overparameterization and produce unrealistic parameter estimates. To address these issues, we propose a novel parameter grouping approach, which reduces the parametric space by co-estimating groups of parameters across compartments. Grouping of parameters is performed using genetic algorithms and is fully automated, based on a novel goodness-of-fit metric. To illustrate the practical application of the proposed methodology, two case studies were conducted. The first case study demonstrates the development of a new PBK model, while the second focuses on model refinement. In the first case study, a PBK model was developed to elucidate the biodistribution of titanium dioxide (TiO2) nanoparticles in rats following intravenous injection. A variety of parameter estimation schemes were employed. Comparative analysis based on goodness-of-fit metrics demonstrated that the proposed methodology yields models that outperform standard estimation approaches, while utilizing a reduced number of parameters. In the second case study, an existing PBK model for perfluorooctanoic acid (PFOA) in rats was extended to incorporate additional tissues, providing a more comprehensive portrayal of PFOA biodistribution. Both models were validated through independent in vivo studies to ensure their reliability.

Venturi Injector Optimization for Precise Powder Transport for Directed Energy Deposition Manufacturing Using the Discrete Element Method and Genetic Algorithms.

Additive manufacturing technologies such as directed energy deposition use powder as their raw material, and it must be deposited in a precise and controlled manner. Venturi injectors could be a solution for the highly precise transport of particulate material. They have been studied from different perspectives, but they are always under high-pressure conditions and mostly fed by gravity. In the present study, an optimization of the different dimensional parameters needed for the manufacturing of a Venturi injector in relation to a particle has been carried out to maximize the amount of powder capable of being sucked and transported for a specific flow in a low-pressure system with high precision in transport. For this optimization, simulations of Venturi usage were performed using the discrete element method, generating different variations proposed by a genetic algorithm based on a preliminary design of experiments. Statistical analysis was also performed to determine the most influential design variables on the objective, with these being the suction diameter (D3), the throat diameter (d2), and the nozzle diameter (d1). The optimal dimensional relationships were as follows: a D3 34 times the particle diameter, a d2 26.5 times the particle diameter, a d1 40% the d2, a contraction angle alpha of 18.73°, and an expansion angle beta of 8.28°. With these proportions, an 85% improvement in powder suction compared to the initial attempts was achieved, with a maximum 2% loss of load.

Integrating fNIRS and machine learning: shedding light on Parkinson's disease detection.

The purpose of this research is to introduce an approach to assist the diagnosis of Parkinson's disease (PD) by classifying functional near-infrared spectroscopy (fNIRS) studies as PD positive or negative. fNIRS is a non-invasive optical signal modality that conveys the brain's hemodynamic response, specifically changes in blood oxygenation in the cerebral cortex; and its potential as a tool to assist PD detection deserves to be explored since it is non-invasive and cost-effective as opposed to other neuroimaging modalities. Besides the integration of fNIRS and machine learning, a contribution of this work is that various approaches were implemented and tested to find the implementation that achieves the highest performance. All the implementations used a logistic regression model for classification. A set of 792 temporal and spectral features were extracted from each participant's fNIRS study. In the two best performing implementations, an ensemble of feature-ranking techniques was used to select a reduced feature subset, which was subsequently reduced with a genetic algorithm. Achieving optimal detection performance, our approach reached 100 % accuracy, precision, and recall, with an F1 score and area under the curve (AUC) of 1, using 14 features. This significantly advances PD diagnosis, highlighting the potential of integrating fNIRS and machine learning for non-invasive PD detection.

A MOGA and extended RPN based approach for heuristic key reliability characteristics analysis in manufacturing.

The manufacturing process is the last opportunity to build an ideal design reliability index into a product. With the advancement of intelligent manufacturing technology, the concept of quality evolves from conformance to fitness for use, which emphasizes that reliability should be built into product with quality control. To effectively implement reliability assurance in the manufacturing process, it is necessary to accurately identify the vital few characteristics that are critical to reliability. Thus, a heuristic key reliability characteristic (KRC) analysis in manufacturing model fusing big quality data is proposed. First, on the basis of the fusion big quality data in manufacturing-by-manufacturing system Reliability-operational process Quality- output product Reliability (RQR) chain, a data driven KRC analysis model is proposed, and a reliability proactive control framework in manufacturing driven by KRC is expounded. Second, considering mass quality and reliability data, an effective KRC identification method based on data mining using multi-objectives genetic algorithm (MOGA) is established. Third, considering manufacturing data and product failure risk, an extended risk priority number (RPN) for KRC ranking is proposed. Finally, an example of an insulating base of subway locomotive is provided to verify the proposed approach.

Engineering mannose-functionalized nanostructured lipid carriers by sequential design using hybrid artificial intelligence tools.

Nanostructured lipid carriers (NLCs) hold significant promise as drug delivery systems (DDS) owing to their small size and efficient drug-loading capabilities. Surface functionalization of NLCs can facilitate interaction with specific cell receptors, enabling targeted cell delivery. Mannosylation has emerged as a valuable tool for increasing the ability of nanoparticles to be recognized and internalized by macrophages. Nevertheless, the design and development of functionalized NLC is a complex task that entails the optimization of numerous variables and steps, making the process challenging and time-consuming. Moreover, no previous studies have been focused on evaluating the functionalization efficiency. In this work, hybrid Artificial Intelligence technologies are used to help in the design of mannosylated drug loaded NLCs. Artificial neural networks combined with fuzzy logic or genetic algorithms were employed to understand the particle formation processes and optimize the combinations of variables for the different steps in the functionalization process. Mannose was chemically modified to allow, for the first time, functionalization efficiency quantification and optimization. The proposed sequential methodology has enabled the design of a robust procedure for obtaining stable mannosylated NLCs with a uniform particle size distribution, small particle size (< 100 nm), and a substantial positive zeta potential (> 20mV). The incorporation of mannose on the surfaces of these DDS following the established protocols achieved > 85% of functionalization efficiency. This high effectiveness should enhance NLC recognition and internalization by macrophages, thereby facilitating the treatment of chronic inflammatory diseases.

Evolutionary Strategies AI Addresses Multiple Technical Challenges in Deep Learning Deployment: Proof-of-Principle Demonstration for Neuroblastoma Brain Metastasis Detection.

Two significant obstacles hinder the advancement of Radiology AI. The first is the challenge of overfitting, where small training data sets can result in unreliable outcomes. The second challenge is the need for more generalizability, the lack of which creates difficulties in implementing the technology across various institutions and practices. A recent innovation, deep neuroevolution (DNE), has been introduced to tackle the overfitting issue by training on small data sets and producing accurate predictions. However, the generalizability of DNE has yet to be proven. This paper strives to overcome this barrier by demonstrating that DNE can achieve satisfactory results in diverse external validation sets. The main innovation of the work is thus showing that DNE can generalize to varied outside data. Our example use case is predicting brain metastasis from neuroblastoma, emphasizing the importance of AI with limited data sets. Despite image collection and labeling advancements, rare diseases will always constrain data availability. We optimized a convolutional neural network (CNN) with DNE to demonstrate generalizability. We trained the CNN with 60 MRI images and tested it on a separate diverse collection of images from over 50 institutions. For comparison, we also trained with the more traditional stochastic gradient descent (SGD) method, with the two variants of (1) training from scratch and (2) transfer learning. Our results show that DNE demonstrates excellent generalizability with 97% accuracy on the heterogeneous testing set, while neither form of SGD could reach 60% accuracy. DNE's ability to generalize from small training sets to external and diverse testing sets suggests that it or similar approaches may play an integral role in improving the clinical performance of AI.

Free-Docking and Template-Based Docking: Physics Versus Knowledge-Based Docking.

Docking methods can be used to predict the orientations of two or more molecules with respect of each other using a plethora of various algorithms, which can be based on the physics of interactions or can use information from databases and templates. The usability of these approaches depends on the type and size of the molecules, whose relative orientation will be estimated. The two most important limitations are (i) the computational cost of the prediction and (ii) the availability of the structural information for similar complexes. In general, if there is enough information about similar systems, knowledge-based and template-based methods can significantly reduce the computational cost while providing high accuracy of the prediction. However, if the information about the system topology and interactions between its partners is scarce, physics-based methods are more reliable or even the only choice. In this chapter, knowledge-, template-, and physics-based methods will be compared and briefly discussed providing examples of their usability with a special emphasis on physics-based protein-protein, protein-peptide, and protein-fullerene docking in the UNRES coarse-grained model.

Growth Characteristics Modeling of Mixed Culture of Bifidobacterium bifidum and Lactobacillus acidophilus using Response Surface Methodology and Artificial Neural Network

Different culture conditions viz. additional carbon and nitrogen content, inoculum size and age, temperature and pH of the mixed culture of Bifidobacterium bifidum and Lactobacillus acidophilus were optimized using response surface methodology (RSM) and artificial neural network (ANN). Kinetic growth models were fitted for the cultivations using a Fractional Factorial (FF) design experiments for different variables. This novel concept of combining the optimization and modeling presented different optimal conditions for the mixture of B. bifidum and L. acidophilus growth from their one variable at-a-time (OVAT) optimization study. Through these statistical tools, the product yield (cell mass) of the mixture of B. bifidum and L. acidophilus was increased. Regression coefficients (R2) of both the statistical tools predicted that ANN was better than RSM and the regression equation was solved with the help of genetic algorithms (GA). The normalized percentage mean squared error obtained from the ANN and RSM models were 0.08 and 0.3%, respectively. The optimum conditions for the maximum biomass yield were at temperature 38°C, pH 6.5, inoculum volume 1.60 mL, inoculum age 30 h, carbon content 42.31% (w/v), and nitrogen content 14.20% (w/v). The results demonstrated a higher prediction accuracy of ANN compared to RSM.

Growth characteristics modeling of Lactobacillus acidophilus using RSM and ANN

The culture conditions viz. additional carbon and nitrogen content, inoculum size, age, temperature and pH of Lactobacillus acidophilus were optimized using response surface methodology (RSM) and artificial neural network (ANN). Kinetic growth models were fitted to cultivations from a Box-Behnken Design (BBD) design experiments for different variables. This concept of combining the optimization and modeling presented different optimal conditions for L. acidophilus growth from their original optimization study. Through these statistical tools, the product yield (cell mass) of L. acidophilus was increased. Regression coefficients (R²) of both the statistical tools predicted that ANN was better than RSM and the regression equation was solved with the help of genetic algorithms (GA). The normalized percentage mean squared error obtained from the ANN and RSM models were 0.06 and 0.2%, respectively. The results demonstrated a higher prediction accuracy of ANN compared to RSM.

Prediction of the mesiodistal size of unerupted canines and premolars for a group of Romanian children: a comparative study

Objectives The aim of the present study was to develop an optimization method of multiple linear regression equation (MLRE), using a genetic algorithm to determine a set of coefficients that minimize the prediction error for the sum of permanent premolars and canine dimensions in a group of young people from a central area of Romania represented by a city called Sibiu. Material and Methods To test the proposed method, we used a multiple linear regression equation derived from the estimation method proposed by Mojers, to which we adjusted regression coefficients using the Breeder genetic algorithm. A total of 92 children were selected with complete permanent teeth with no clinically visible dental caries, proximal restorations or orthodontic treatment. A hard dental stone was made for each of these models, which was then measured with a digital calliper. The Dahlberg analyses of variance had been performed to determine the error of method, then the Correlation t Test was applied, and finally the MLRE equations were obtained using the version 16 for Windows of the SPSS program. Results The correlation coefficient of MLRE was between 51-67% and the significance level was set at α=0.05. Comparing predictions provided by the new and respectively old method, we can conclude that the Breeder genetic algorithm is capable of providing the best values for parameters of multiple linear regression equations, and thus our equations are optimized for the best performance. Conclusion The prediction error rates of the optimized equations using the Breeder genetic algorithm are smaller than those provided by the multiple linear regression equations proposed in the recent study. .

METODOLOGÍA PARA DIMENSIONAMIENTO DE MECANISMO POLICÉNTRICO DE RODILLA UTILIZANDO ANÁLISIS DE MARCHA Y ALGORITMOS GENÉTICOS / METHODOLOGY TO GAUGE A FOUR-BAR LINKAGE PROSTHETIC KNEE MECHANISM BASED ON GAIT ANALYSIS AND GENETIC ALGORITHMS

El objetivo de esta investigación es desarrollar una metodología para dimensionar un mecanismo policéntrico de rodilla de 4 barras para máxima estabilidad. Basado en el hecho de que la estabilidad del mecanismo durante la respuesta a la carga depende de la posición del centro instantáneo de rotación (CIR) respecto la fuerza de reacción del piso (FRP) durante la fase de apoyo, se desarrolló una plataforma de cómputo que representa el movimiento real de la pierna, el vector FRP y el mecanismo con su CIR. Para obtener los datos de entrada a la plataforma, se realizó un análisis de marcha a una paciente con amputación transfemoral unilateral, obteniendo la FRP, el ángulo de flexo-extensión de rodilla y la cinemática de los miembros inferiores. Por otra parte, a través de los algoritmos genéticos (AGs), se obtienen las dimensiones y configuración de los eslabones del mecanismo requeridas para iterar con la plataforma en la cual, comparando la ubicación de la FRP respecto al CIR en el plano sagital, se determinan las dimensiones funcionales adecuadas. El mecanismo se dimensionó exitosamente utilizando la metodología desarrollada, garantizando estabilidad de la rodilla después del contacto inicial y flexión voluntaria antes del despegue de punta.

This research was aimed to develop a methodology for establishing the proper dimensions of a four-bar linkage prosthetic knee mechanism for maximum stability. Based on the fact that the stability of a four-bar knee during load-bearing is determined by the location of the instantaneous center of rotation (ICR) with respect to the ground reaction force (GRF) vector, a computational platform was developed to simulate the movement of the leg, the GRF vector and the position of the ICR of the mechanism. On one hand, a gait analysis was carried out on a subject with unilateral transfemoral amputation, from which the GRF, the knee flexion-extension angle and the kinematics of the lower limbs were determined. On the other hand, genetic algorithms (GAs) technique provided the dimensions and mechanism links configuration required to iterate with the platform on which, comparing the location of the GRF and the ICR in the sagittal plane, the functional dimensions of the mechanism were obtained. The polycentric knee mechanism was gauged successfully by ensuring knee stability during the initial contact and load response as well as the ability to initiate voluntary flexion toward late stance before the toe-off.

Algoritmos Genéticos aplicados a la Ingeniería biomédica / Genetic algorithms applied to Biomedical Engineering

En este trabajo se describe las aplicaciones y alcances del método de los algoritmos genéticos (AG) en la investigación en bioingeniería, mecanobiología y medicina. Para este fin, se ha desarrollado el trabajo sobre tres artículos recientes que describen las aplicaciones de los AG en problemas de ingeniería biomédica. Este trabajo pone de manifiesto la importancia del uso de nuevas metodologías de optimización en las investigaciones biomédicas.

In present paper are described the applications and scope of the genetic algorithms method (GA) in the case of the research in the bioengineering, mechanobiology and medicine. For this aim, the paper on three recent articles was developed describing the applications of the GA in problems related to biomedical engineering. Present paper emphasizes the significance of the use of new methodologies of optimization in the biomedical researches.

Growth characteristics modeling of Bifidobacterium bifidum using RSM and ANN

The aim of this work was to optimize the biomass production by Bifidobacterium bifidum 255 using the response surface methodology (RSM) and artificial neural network (ANN) both coupled with GA. To develop the empirical model for the yield of probiotic bacteria, additional carbon and nitrogen content, inoculum size, age, temperature and pH were selected as the parameters. Models were developed using » fractional factorial design (FFD) of the experiments with the selected parameters. The normalized percentage mean squared error obtained from the ANN and RSM models were 0.05 and 0.1 percent, respectively. Regression coefficient (R²) of the ANN model showed higher prediction accuracy compared to that of the RSM model. The empirical yield model (for both ANN and RSM) obtained were utilized as the objective functions to be maximized with the help of genetic algorithm. The optimal conditions for the maximal biomass yield were 37.4 °C, pH 7.09, inoculum volume 1.97 ml, inoculum age 58.58 h, carbon content 41.74 percent (w/v), and nitrogen content 46.23 percent (w/v). The work reported is a novel concept of combining the statistical modeling and evolutionary optimization for an improved yield of cell mass of B. bifidum 255.

Linear versus non-linear noninvasive temperature prediction in a homogeneous medium subjected to physiotherapeutic ultrasound / Comparação entre predição linear e não linear não-invasiva de temperatura num meio homogêneo sujeito a ultra-som de terapia

The lack of accurate time-spatial temperature estimators/predictors conditions the safe application of thermal therapies, such as hyperthermia. In this paper, a comparison between a linear and a non-linear class of models for non-invasive temperature prediction in a homogeneous medium, subjected to ultrasound at physiotherapeutic levels is presented. The linear models used were autoregressive with exogenous inputs (ARX) and the non-linear models were radial basis functions neural networks (RBFNN). In order to create and validate the models, an experiment was build to extract in vitro ultrasound RF-lines, as well as its correspondent temperature values. Then, features were extracted from the measured RF-lines and the models were trained and validated. For both the models, the best-fitted structures were selected using the multi-objective genetic algorithm (MOGA), given the enormous number of possible structures. The best RBFNN model presented a maximum absolute predictive error in the validation set five times less than the value presented by the best ARX model. In this work, the best RBFNN reached a maximum absolute error of 0.42 ºC, which is bellow the value pointed as a borderline between an appropriate and an undesired temperature estimator, which is 0.5 ºC. The average error was one order of magnitude less in the RBFNN case, and a less biased estimation was met. In addition, the best RBFNN needed less environmental information(inputs), given the capacity to non-linearly relate the information. The results obtained are encouraging, considering that coherent results should be obtained in a time-spatial modelling schema using RBFNN models.

A falta de estimadores de temperatura espaço-temporais que sejam precisos impede a aplicação segura das terapias térmicas, como por exemplo a hipertermia. Neste artigo é apresentada uma comparação entre uma classe de modelos lineares e uma classe de modelos não lineares, na predição não invasiva de temperatura num meio homogêneo, quando o mesmo é aquecido por ultra-som em níveis usados em fisioterapia. Os modelos lineares considerados foram do tipo auto-regressivo com entradas exógenas (ARX); a nível não-linear foram considerados redes neuronais RBF (RBFNN). Para treinar e validar os modelos foram recolhidas os ecos provenientes do meio, bem como os correspondentes valores de temperatura. Após a colheita de informação, foram extraídas características dos ecos medidos e posteriormente os modelos foram treinados e validados. Para ambas as classes de modelos, as melhores estruturas foram seleccionadas usando um algoritmo genético multi-objectivo (MOGA), devido ao número elevado de estruturas possíveis. O melhor modelo RBFNN apresentou um erro máximo absoluto cinco vezes inferior ao erro máximo absoluto apresentado pelo melhor modelo ARX. Neste trabalho, o melhor modelo RBFNN apresentou um erro máximo absolutode 0,42 ºC, valor este que é inferior ao limite (0,5 ºC) apresentado como sendo a fronteira entre um estimador desejado e um estimador indesejado. O erro médio cometido pelo melhor modelo neuronal é uma ordem de grandeza inferior ao erro médio apresentado pelo melhor modelo linear, obtendo-se deste modo uma estimação menos enviesada no caso das redes neuronais, com menos informação do ambiente (menos entradas) devido ao processamento não-linear dos dados de entrada. Os resultados obtidos são encorajadores, apontando no sentido de se obter bons resultados numa estimação espaço-temporal.