Nowadays, colorectal cancer is one of the most common cancers, and early detection would greatly help improve patient survival. The current methods used by physicians to detect it are based on the visual detection of polyps in colonoscopy, a task that can be tackled by means of semantic segmentation methods. However, the amount of data necessary to train deep learning models for these problems is a barrier for their adoption. In this work, we study the application of different semi-supervised learning techniques to this problem when we have a small amount of annotated data. In this study, we have used the Kvasir-SEG data set, taking only 60 and 120 annotated images and studying the behaviour of the Data Distillation, Model Distillation, and Data & Model distillation methods in both cases, using 10 different architectures. The results show that as we increase the number of initially annotated data, most models obtained better results, but two of them performed worse in the baseline case. Furthermore, we can conclude that the Data Distillation method increases the performance of the models a 48.6% and 30.6% on average using 60 and 120 annotated images respectively. Finally, using only 12% of the annotated data and applying Data Distillation, the results obtained are not very far from those obtained by training the models with the fully annotated dataset. For all these reasons, we conclude that the Data Distillation method is a good tool in semantic segmentation problems when the number of initially annotated images is small.
This study explores prodromal Parkinson’s Disease (PD) by leveraging data from the Parkinson’s Progression Markers Initiative (PPMI). The main goal was to discriminate between prodromals that phenoconverted to PD in 7 years to those that did not. Through feature selection, the system identified key first visit predictors of PD phenoconversion, encompassing demographic, clinical, and structural magnetic resonance imaging (MRI) data. Employing seven machine learning algorithms in standard and balanced forms, we find Support Vector Machine (balanced) as most effective for demographic and clinical data, and Logistic Regression (balanced) when adding thicknesses and volumes of MRI data. The metrics were improve in the second case (AUC ROC of 0.84). Significant predictors include olfactory dysfunction, motor symptoms, psychomotor speed, and third ventricle dilation.
This paper contributes with an alternative to the multivariate Analogue Method (AM) version, using a preprocessing stage carried out by an Autoencoder (AE). The proposed method (MvAE-AM) is applied to reconstruct France’s 2003, Balkans’ 2007 and Russia 2010 mega heat waves. Using divers such as geopotential height of the 500hPA (Z500), mean sea level pressure (MSL), soil moisture (SM), and potential evaporation (PEva), the AE extracts the most relevant information into a smaller univariate latent space. Then, the classic univariate AM is applied to search for similar situations in the past over the latent space, with a minimum distance to the heat wave under evaluation. We have compared the proposed method’s performance with that of a classical multivariate AM (MvAM), showing that the MvAE-AM approach outperforms the MvAM in terms of accuracy ( + 1.1257 C), while reducing the problem’s dimensionality.
E-commerce has become an essential aspect of modern life, providing consumers globally with convenience and accessibility. However, the high volume of short and noisy product descriptions in text streams of massive e-commerce platforms translates into an increased number of clusters, presenting challenges for standard model-based stream clustering algorithms. Standard LDA-based methods often lead to clusters dominated by single elements, effectively failing to manage datasets with varied cluster sizes. Our proposed Community-Based Topic Modeling with Contextual Outlier Handling (CB-TMCOH) algorithm introduces an approach to outlier detection in text data using transformer models for similarity calculations and graph-based clustering. This method efficiently separates outliers and improves clustering in large text datasets, demonstrating its utility not only in e-commerce applications but also proving effective for news and tweets datasets.
Embryo selection is an indispensable step to ensure the success of in vitro fertilization. There are two techniques to perform embryo selection: preimplantation genetic screening and embryo morphological grading. However, even with these techniques, the embryo implantation probability is barely 65% making extremely difficult to evaluate their implantation potential. This is mainly due to the lack of markers, and the subjectivity associated with experience, judgment, and training of the embryologists. Computer vision and deep learning methods can help to automatically identify those markers with methods such as the segmentation of the embryo structures to offer detailed, quantitative, and objective assessments; and with that, information to predict the pregnancy outcome of embryos. In this paper, we present different methods capable of segmenting the components of an embryo (namely, the Trophectoderm, the Inner Cell Mass and the Zona Pellucida) with Dice scores ranging from 0.85 to 0.89, and openly release the code so that anyone can use it and replicate the results. These models are a first step towards a more objective evaluation of the embryos’ implantation potential.
