Breast cancer is the most common cancer in women. While mammography is the most widely used screening technique for the early detection of this disease, it has several disadvantages such as radiation exposure or high economic cost. Recently, multiple authors studied the ability of machine learning algorithms for early diagnosis of breast cancer using thermal images, showing that thermography can be considered as a complementary test to mammography, or even as a primary test under certain circumstances. Moreover, although some personal and clinical data are considered risk factors of breast cancer, none of these works considered that information jointly with thermal images. We propose a novel approach for early detection of breast cancer combining thermal images of different views with personal and clinical data, building a multi-input classification model which exploits the benefits of convolutional neural networks for image analysis. First, we searched for structures using only thermal images. Next, we added the clinical data as a new branch of each of these structures, aiming to improve its performance. We applied our method to the most widely used public database of breast thermal images, the Database for Mastology Research with Infrared Image. The best model achieves a 97% accuracy and an area under the ROC curve of 0.99, with a specificity of 100% and a sensitivity of 83%. After studying the impact of thermal images and personal and clinical data on multi-input convolutional neural networks for breast cancer diagnosis, we conclude that: (1) adding the lateral views to the front view improves the performance of the classification model, and (2) including personal and clinical data helps the model to recognize sick patients.

Breast cancer is the most common cancer in women. Ultrasound is one of the most used techniques for diagnosis, but an expert in the field is necessary to interpret the test. Computer-aided diagnosis (CAD) systems aim to help physicians during this process. Experts use the Breast Imaging-Reporting and Data System (BI-RADS) to describe tumors according to several features (shape, margin, orientation...) and estimate their malignancy, with a common language. To aid in tumor diagnosis with BI-RADS explanations, this paper presents a deep neural network for tumor detection, description, and classification. An expert radiologist described with BI-RADS terms 749 nodules taken from public datasets. The YOLO detection algorithm is used to obtain Regions of Interest (ROIs), and then a model, based on a multi-class classification architecture, receives as input each ROI and outputs the BI-RADS descriptors, the BI-RADS classification (with 6 categories), and a Boolean classification of malignancy. Six hundred of the nodules were used for 10-fold cross-validation (CV) and 149 for testing. The accuracy of this model was compared with state-of-the-art CNNs for the same task. This model outperforms plain classifiers in the agreement with the expert (Cohen’s kappa), with a mean over the descriptors of 0.58 in CV and 0.64 in testing, while the second best model yielded kappas of 0.55 and 0.59, respectively. Adding YOLO to the model significantly enhances the performance (0.16 in CV and 0.09 in testing). More importantly, training the model with BI-RADS descriptors enables the explainability of the Boolean malignancy classification without reducing accuracy.

Social networks allow us to communicate with people around the world. However, some users usually take advantage of anonymity for writing offensive comments to others, which might affect those who receive offensive messages or discourage the use of these networks. However, it is impossible to manually check every message. This has promoted several proposals for automatic detection systems. Current state-of-the-art systems are based on the transformers’ architecture and most of the work has been focused on the English language. However, these systems do not pay too much attention to the unbalanced nature of data, since there are fewer offensive comments than non-offensive in a real environment. Besides, these previous works have not studied the impact on the final results of pre-processing or the corpora used for pre-training the models. In this work, we propose and evaluate a series of automatic methods aimed at detecting offensive language in Spanish texts addressing the unbalanced nature of data. We test different learning models, from those based on classical Machine Learning algorithms using Bag-of-Words as data representation to those based in large language models and neural networks such as transformers, paying more attention to minor classes and the corpora used for pre-training the transformer-based models. We show how transformer-based models continue obtaining the best results, but we improved previous results by a 6,2% by adding new steps of pre-processing and using models pre-trained with Spanish social-media data, setting new state-of-the-art results.

Recent research has shown that Massive Open Online Courses (MOOCs) create barriers for students with disabilities. Not taking into account their needs in the design, production or delivery of MOOCs may be one of the main causes behind this. It leads to poor compliance with suitable learning designs and web accessibility standards, as well as a lack of knowledge about the students’ needs. The objective of our research is to analyze the learning performance of the students in MOOCs on topics related to Design for All, offered in an Open edX-based platform. Accessibility support was conceived from the outset, including compliance of both the platform and the learning resources with the WCAG 2.1 accessibility standard, and with a subset of the principles of Universal Design for Learning. Additionally, students were consulted on their accessibility needs and preferences, following publicly available modeling schemes and previous research. From a sample of 765 students, who completed at least one of the graded assessment activities of the course, a multilevel multiple logistic regression model was fitted. Based on that model, the results indicate that: a) users of screen readers and users of captions show a statistically significant positive association with a good performance when compared to students with no preferences, with an odds ratio of, respectively, OR = 13.482 and OR = 13.701; b) students who have low vision or very low vision show a significant negative association with a good performance when compared to users of screen readers and to users of captions, with OR = 26.817 and OR = 27.254, respectively.

The study and early detection of breast cancer are key for its treatment. We carry out an exhaustive analysis of the most used database for mastology research with infrared images, analyzing the anomalies according to five quality dimensions: completeness, correctness, concordance, plausibility, and currency. We established control queries that looked for these anomalies and that can be used to ensure the quality of the database. Finally, we briefly review the more than 40 papers that use this database and that do not mention any of these anomalies. When analyzing the database, we found 365 anomalies related to personal and clinical data, and thermal images. The errors found in our research may lead to a modification of the results and conclusions made in the articles found in the literature, serve as a basis for improvements in the quality of the database, and help future researchers to work with it.