Modeling tumor progression from mammographic images remains a challenging and inherently ill-posed problem (because multiple valid outcomes or plausible futures exist), particularly in the absence of longitudinal datasets. In this work, we propose a conditional Generative Adversarial Network (cGAN) framework to model statistically plausible tumor progression patterns directly in the image domain, without assuming an explicit temporal horizon. Given a region of interest (ROI) containing a tumor, the model is trained to generate a corresponding future-appearance ROI, conditioned on the input image. Because true longitudinal pairs are unavailable in the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM), we construct clinically consistent pseudo-pairs by matching lesions according to radiological attributes and requiring a minimum size increase criterion. We validate this pairing strategy through expert radiological review. Our training objective combines adversarial loss with reconstruction-based terms, along with stabilization techniques such as instance noise and label smoothing. Quantitative evaluation using LPIPS and SSIM metrics demonstrates structural and perceptual coherence between generated projections and the reference targets, while qualitative assessment by a licensed radiologist indicates that the majority of generated samples are clinically plausible. Although preliminary, and based on pseudo-longitudinal data, these results suggest that cGANs provide a viable data-driven alternative to explicit tumor growth modeling, opening new directions for longitudinal analysis and decision support in mammographic imaging.

A well-known problem in Computer Vision is the detection of interest points, which is important for tasks such as object detection, image matching, object recognition, 3D reconstruction, tracking, and image registration. This work proposes the use of Grammatical Evolution to perform the automated design of Interest Point Detectors, in search of the optimum detector, for image matching. Interest Point Detectors were evolved with base on low level operations such as Gaussian filters, Laplacian filters, etc. and intensity operations. Images with planar transformations (rotation, translation, and scaling), that simulate homographies between the original and transformed images, were used to guide the evolutionary search, based on the repeatability rate computed on the detected points. With repeatability rates generally above 80

BACKGROUND AND OBJECTIVE:In the Tumor Micro-Environment, cancer progression and its relationship with the Immune System (IS) are described in terms of cancer immunoediting (CI) phases, each of which is characterized by different types and levels of interaction between the tumor cells and elements of the IS, such as CD8+T cells. Said interactions are governed by genotypical (Tumor Mutational Burden, TMB) and phenotypical aspects pertaining to the tumor, as well as by the strength of the IS. In this work, a computational model of CI is presented that incorporates the TMB and the biomarker Tumor Immunogenic Phenotype (TIP) as its control parameters, and which employs the Ising-model Hamiltonian to characterize the system with respect to the CI phases. METHODS:Our model is a probabilistic multi-agent system with agents for tumor cells and for the IS. The computer implementation includes the parametrization of the TMB and the TIP, which is useful for identifying whether a tumor is hot or cold based on tumor immunogenicity and inflammation. For modeling the interactions between tumor and immune cells, the relevant elements are integrated under a Michaelis-Menten equation that regulates the recruitment rate of CD8+T cells and other IS elements. This novel quantification of immune cell recruitment encompasses the growth of neoantigen production, which in turn triggers the growth of CD8+T cells. RESULTS:Our model reliably captures the Elimination, Equilibrium, and Escape phases of tumor-immune cell interactions, modulating the observed behaviors through the introduced parametrization of TMB and TIP biomarkers. Notably, these results align well with the combination of genotypical and phenotypical biomarkers analyzed in recent literature. A remarkable instance is the appreciable inhibition of the tumor activity during the Escape phase, observed for phenotypically hot tumors with relatively high TMB, and pointing towards improved efficacy of the IS against such tumors. The Ising-Hamiltonian provides precise quantification of diverse tumor-immune interactions across different TMB and TIP value combinations. CONCLUSIONS:The presented model, formed by relatively simple agents, generates emergent behaviors through which the phases of CI are identified. The flexible choice of control parameters is robust enough and provides a plausible explanation for the mechanisms through which tumors with high TMB and high immunogenicity (i.e., hot tumors) exhibit a higher probability of responding to immunotherapy treatment. Characterization via the Ising-model Hamiltonian supports this explanation by summarizing the system's dynamics, which, in turn, facilitates its analysis and methodical improvements. The complex interplay of TMB, TIP, and individual physiology is finely captured.

Background and Objective: In the Tumor Micro-Environment, cancerprogression and its relationship with the Immune System (IS) is described interms of cancer immunoediting (CI) phases, each of which is characterized bydifferent types and levels of interaction between the tumor cells and elements ofthe IS, such as CD8+T cells. Said interactions are governed by genotypical (TumorMutational Burden, TMB) and phenotypical aspects pertaining to the tumor, aswell as by the strength of the IS. In this work, a computational model of CI ispresented which incorporates the TMB and the biomarker Tumor Immunogenic Phenotype (TIP) as its control parameters, and which employs the Ising-modelHamiltonian to characterize the system with respect to the CI phases. Methods: Our model is a probabilistic multi-agent system with agents for tumorcells and for the IS. A basic version of this model was presented before; for thiswork we have produced a new implementation of our system, improved throughthe inclusion of the TMB and the TIP (i.e. whether a tumor is hot or cold). Thenew elements are integrated under a Michaelis-Menten relationship that isemployed to regulate the recruitment rate of CD8+T cells (and other IS elements),thus controlling the interactions between tumor cells and the IS. Results: Our simulations confirm that the proposed system is capable ofconsistently generating different phases of CI, and that by varying the introducedparameters, the system is effectively controlled. However, beyond the expectedbehaviors, we also found unanticipated effects that nevertheless match well withthe literature regarding the combination of the genotypical and phenotypicalbiomarkers discussed. Moreover, the Ising-model Hamiltonian is confirmed as avaluable tool for the broad characterization of tumor-IS interaction, by reflectingclear and distinct patterns related to the phases of CI. Conclusions: The presented model, although formed by relatively simple agents,generate emergent behaviors with which the phases of CI can be identified. Themodel is robust to the choice of its control parameters and, more importantly,provides a plausible explanation for the mechanics through which tumors withhigh TMB and high immunogenicity (i.e. hot tumors) exhibit a higher probabilityof responding to treatment by immunotherapy. Characterization via theIsing-model Hamiltonian also contributes to said explanation, by summarizing thesystem’s dynamics in a way that facilitates its analysis and future improvements.

In real-world classification tasks, it is common to find class imbalance issues in the training datasets, i.e. an unequal number of examples among the different classes. The class imbalance problem biases the performance of predictive models by overlooking minority classes; this is because predictive models employ learning rules with accuracy-based cost functions, thus favoring majority classes. In this work, the class imbalance issue is tackled through generative models, using the BreakHis dataset, a histopathologic image set intended for breast cancer classification, as a case study. The BreasHis' minority class is balanced by adding synthetic images obtained by means of different generative methods, including variational autoencoders and two different generative adversarial networks. The quality of the image sets created by the different generative models, and their effects in balancing the BreakHis dataset, are evaluated through several quantitative metrics computed from classification tasks. Statistical analysis is performed and the results indicate that the DCGAN network is superior to the other evaluated models.

The advent of Deep Learning introduced a paradigm shift in the design and implementation of machine learning models, from the feature engineering paradigm towards the feature learning one; nowadays much less effort is dedicated to the manufacture of feature extraction methods, albeit at the expense of requiring larger volumes of training data and extended training times for deep models to learn the meaningful features. Nonetheless, it has been observed that the initial layers of many image models tend to converge to some of the earlier engineered feature extractors, mainly in the form of Gabor filters and other spatial filters, thus generating a growing interest in replacing the first layers in CNNs with learnable spatial filters, and in particular, Gabor filters. In this work we investigate the problem of parameter convergence in learnable Gabor filters, discover that the filters can exhibit degradation after a few epochs of training, and propose a method based on similarity metrics between the Gabor filters to address this issue. This research can contribute to the design of more efficient training strategies of networks employing learnable spatial filters, to leverage their intrinsic advantages over the more popular non-engineered convolutional filters.

Grammar Evolution (GE) represents a variant of Genetic Programming that has gained substantial popularity in recent years within the field of Automatic Programming. Extensive research has been conducted on various aspects of GE, including its core components such as the Search Engine, the Mapping Process, and the employed Grammar. Nevertheless, a noteworthy recent development centers around the optimization of codon selection methods. This paper presents a novel methodology aimed at enhancing codon selection through the utilization of permutation strategies. The application of this methodology is demonstrated in the context of addressing the symbolic regression problem, and the obtained results are subjected to rigorous scrutiny through statistical tests.

The Automated Machine Learning (AutoML) process for feature selection, model creation and hyper-parameter optimization was performed. The AutoML process is applied to generate a machine learning model for the forecasting of Google’s stock prices. For this reason the feature selection task is performed based on a set of 11 technical indicators, including the time series of the stock prices. Model selection is carried out between two models, a Long-Short Term Memory (LSTM) by itself or combined with Convolutional Neural Network (CNN). Hyper-parameter optimization is done on the learning rate, the windows size, the batch size, the epochs, the number of CNN filters and the CNN’s kernel size. To perform the AutoML process, a bio-inspired population metaheuristic, Particle Swarm Optimization (PSO), is used. The experimental results show that the errors of the AutoML-generated solutions are in the order of 1E-2 to 9E-4 and that a CNN-LSTM network has better performance than an LSTM network by itself.

Resumen. El proceso de optimizaci´on de hiperpar´ametros de una red neuronal artificial, usualmente se lleva cabo de manera manual usando Grid Search o Random Search. En este art´ıculo, estudiamos una metodolog´ıa para optimizar hiperpar´ametros de una red Long Short Term Memory (LSTM) usando dos me-taheur´ısticas bio-inspiradas, Particle Swarm Optimization, y Flower Pollination Algorithm y una Estimation of Distribution Algorithm (EDA), Low Number of Function Evaluation, para que la red realice un pron´ostico financiero del precio de cierre del siguiente d´ıa de las acciones de Google y Nike. Los resultados muestran que las redes LSTM obtenidas mediante el proceso de optimizaci´on de hiperpar´ametros son m´as simples, resultando en menor tiempo de entrenamiento, y mayor rendimiento, que las redes no optimizadas con metaheur´ısticas. Los errores de prueba obtenidos de las soluciones son de 10 E-4 hasta 10 E-6. Adi-cionalmente realizamos una comparaci´on de las metaheur´ısticas, concluyendo que la EDA encuentra una soluci´on esperada con menor cantidad de llamadas a funci´on, lo que se traduce en menor costo computacional y

An intelligent system intended to provide assistance to an elderly user in their home is presented. This multi-agent system is designed to monitor the interaction between the user and other agents, and regulate itself based on the estimated level of mental engagement of the user in order to contribute to help them maintain adequate levels of awareness and interaction with their home environment. The agents are devices programmed to preserve the security, comfort and health of the user, who is also conceptualized as one of the agents in the system. The user engagement level is estimated through fuzzy inference, that is, by means of a set of fuzzy membership functions and rules applied over the interaction of the user with different categories of agents. The system incorporates several dozen agents and is tested using two different datasets of real interactions between users and devices in their home environments. A uniform parameter configuration yielded very positive results for diverse scenarios and subsequently the parameters of the system were automatically optimized to show that improved performance, in the form of a more efficient system regulation and increased certainty in adjustment estimation, is obtained.

Background and objective Cancer Immunoediting (CI) describes the cellular-level interaction between tumor cells and the Immune System (IS) that takes place in the Tumor Micro-Environment (TME). CI is a highly dynamic and complex process comprising three distinct phases (Elimination, Equilibrium and Escape) wherein the IS can both protect against cancer development as well as, over time, promote the appearance of tumors with reduced immunogenicity. Herein we present an agent-based model for the simulation of CI in the TME, with the objective of promoting the understanding of this process. Methods Our model includes agents for tumor cells and for elements of the IS. The actions of these agents are governed by probabilistic rules, and agent recruitment (including cancer growth) is modeled via logistic functions. The system is formalized as an analogue of the Ising model from statistical mechanics to facilitate its analysis. The model was implemented in the Netlogo modeling environment and simulations were performed to verify, illustrate and characterize its operation. Results A main result from our simulations is the generation of emergent behavior in silico that is very difficult to observe directly in vivo or even in vitro . Our model is capable of generating the three phases of CI; it requires only a couple of control parameters and is robust to these. We demonstrate how our simulated system can be characterized through the Ising-model energy function, or Hamiltonian, which captures the “energy” involved in the interaction between agents and presents it in clear and distinct patterns for the different phases of CI. Conclusions The presented model is very flexible and robust, captures well the behaviors of the target system and can be easily extended to incorporate more variables such as those pertaining to different anti-cancer therapies. System characterization via the Ising-model Hamiltonian is a novel and powerful tool for a better understanding of CI and the development of more effective treatments. Since data of CI at the cellular level is very hard to procure, our hope is that tools such as this may be adopted to shed light on CI and related developing theories.

Fostered by technological and theoretical developments, deep neural networks (DNNs) have achieved great success in many applications, but their training via mini-batch stochastic gradient descent (SGD) can be very costly due to the possibly tens of millions of parameters to be optimized and the large amounts of training examples that must be processed. The computational cost is exacerbated by the inefficiency of the uniform sampling typically used by SGD to form the training mini-batches: since not all training examples are equally relevant for training, sampling these under a uniform distribution is far from optimal, making the case for the study of improved methods to train DNNs. A better strategy is to sample the training instances under a distribution where the probability of being selected is proportional to the relevance of each individual instance; one way to achieve this is through importance sampling (IS), which minimizes the gradients' variance w.r.t. the network parameters, consequently improving convergence. In this paper, an IS-based adaptive sampling method to improve the training of DNNs is introduced. This method exploits side information to construct the optimal sampling distribution and is dubbed regularized adaptive sampling (RAS). Experimental comparison using deep convolutional networks for classification of the MNIST and CIFAR-10 datasets shows that when compared against SGD and against another sampling method in the state of the art, RAS produces improvements in the speed and variance of the training process without incurring significant overhead or affecting the classification.

Training of Deep Neural Networks (DNNs) is very computationally demanding and resources are typically spent on training-instances that do not provide the most benefit to a network’s learning; instead, the most relevant instances should be prioritized during training. Herein we present an improved version of the Adaptive Sampling (AS) method (Gopal, 2016) extended for the training of DNNs. As our main contribution we formulate a probability distribution for data instances that minimizes the variance of the gradient-norms w.r.t. the network’s loss function. Said distribution is combined with the optimal distribution for the data classes previously derived by Gopal and the improved AS is used to replace uniform sampling with the objective of accelerating the training of DNNs. Our proposal is comparatively evaluated against uniform sampling and against Online Batch Selection (Loshchilov & Hutter, 2015). Results from training a Convolutional Neural Network on the MNIST dataset with the Adadelta and Adam optimizers over different training batch-sizes show the effectiveness and superiority of our proposal.

In this work, we focus on the problem of selecting low-level heuristics in a hyperheuristic approach with offline learning, for the solution of instances of different problem domains. The objective is to improve the performance of the offline hyperheuristic approach, identifying equivalence classes in a set of instances of different problems and selecting the best performing heuristics in each of them. A methodology is proposed as the first step of a set of instances of all problems, and the generic characteristics of each instance and the performance of the heuristics in each one of them are considered to define the vectors of characteristics and make a grouping of classes. Metalearning with statistical tests is used to select the heuristics for each class. Finally, we used the Naive Bayes to test the set instances with k-fold cross-validation, and we compared all results statistically with the best-known values. In this research, the methodology was tested by applying it to the problems of capacitated vehicle routing (CVRP) and graph coloring (GCP). The experimental results show that the proposed methodology can improve the performance of the offline hyperheuristic approach, correctly identifying the classes of instances and applying the appropriate heuristics in each case. This is based on the statistical comparison of the results obtained with those of the state of the art of each instance.

We present a device designed to provide assistance to an older adult in their home. Conceptually, this assistant is conceived as a multi-agent system designed to monitor and regulate itself according to the mental engagement of the user and contribute to help them maintain adequate levels. Physically, the assistant is realized as a desk, equipped with an automated pill dispenser and interconnected with other devices intended to provide security and comfort to the user. Through simulations it was possible to confirm that the system performs as desired for diverse scenarios. This conclusion encourages us to proceed with the next stage of development, where real data will be collected employing the prototype of the system described herein.

. Breast cancer is the most common form of cancer in the female population. As with any form of cancer, early detection of breast cancer is one of the most important factors affecting the possibility of recovery from the disease. Early detection of breast cancer can be achieved through mammography screening programs. Studies have shown that double reading of mammograms improves the detection of breast abnormalities. Unfortunately, the enormous amount of mammograms to be examined prohibits the practice of double reading by human experts. Computerized systems for automated detection and classification of breast abnormalities in mammograms have been developed as a possibility to alleviate this problem. These systems have the objective of accompanying human experts and eventually replace one of them in double or triple reading of mammograms. Some of the signs associated with breast cancer that can be observed in mammograms include: masses, calcifications, distortion of the parenchimal tissue, asymmetry of breast tissue between the breasts of a patient, etc. Beginning from about three decades ago, computerized systems and algorithms specialized in the detection and diagnosis of each type of these signs have been proposed by researchers and described in the scientific literature. Interest in the problem persists to this day; the development of automated systems for detection/diagnosis of breast cancer is currently a very active research area. In this work, an overview of the techniques developed for diagnosis of breast cancer is presented from a perspective of the work that has been preciously carried out personally in this area.

Effective computer applications targeting healthcare problems can be traced back to at least the second half of the 20th century. During the first half of the 21st century, this area of application has experienced a revolution, driven by the advent and development of advanced Machine Learning techniques, and supported by the ever more powerful computer hardware, electronics, and the increasingly prevalent global-communications infrastructure. All of these are bringing the medical fields, and computer science, closer together than they have ever been. This thematic section collects research derived from the current intense interaction between computer scientists working on artificial intelligence, machine learning and big data, and clinicians / medical researchers working on the different branches of health care. The field of MLHC supports the advancement of data analytics, knowledge discovery, and the meaningful use of complex medical data by fostering collaboration and the exchange of ideas between these communities. The world health challenges on cancer, malaria and retinal diseases growth, that involve intense data analysis as well as precise but flexible modeling, concern this volume of AI solutions. In “CAD of Breast Cancer: a decade-long review of techniques for Mammography Analysis”, Rojas et al. discuss the paradigm shift observed within the machine learning research community (from feature-engineering to feature-learning) and how it is reflected on the design of computer-aided systems for mammography analysis used against breast cancer. In “Machine Learning Techniques for Diagnosis of Breast Cancer”, A. Rojas offers an overview of the recent machine learning techniques for diagnosis of breast cancer, presented from a perspective of the work that he has carried out through several years. The techniques used in the two primary tasks for the early diagnosis of breast cancer (mass detection and mass classification) are discussed. In “Evaluation of breast cancer by infrared thermography”, Morales-Cervantes et al., discuss the analysis of thermograms of patients with suspected breast cancer. The asymmetry of a thermal score (combination of the amount of vascularization and surface temperature) between the breasts of a patient is indicative of anomalies. The automated method presented achieves higher sensitivity (100%) and specificity (68.68%) than an expert oncologist, on 206 test thermograms. In “Cancer metastasis and the immune system response: CM-IS modeling by Ising model”, Alvarado and Arroyo introduce this highly complex biological process, of top interest for cancer diagnosis and therapy. The strength of the immune system response against cancer correlates with the success of the cancer growth. Their interaction is formalized by the Ising model (a classic model for emergent-interaction phenomena) and simulated by means of an agent-based environment. In “Automatic Cropping of Retinal Fundus Photographs using Convolutional Neural Networks”, González-Briceño et al. present a segmentation method based on Deep