Remote sensing has emerged as a critical tool for large-scale Earth monitoring and land management. In this paper, we introduce AgriPotential, a novel benchmark dataset composed of Sentinel-2 satellite imagery captured over multiple months. The dataset provides pixel-level annotations of agricultural potentials for three major crop types - viticulture, market gardening, and field crops - across five ordinal classes. AgriPotential supports a broad range of machine learning tasks, including ordinal regression, multi-label classification, and spatio-temporal modeling. The data cover diverse areas in Southern France, offering rich spectral information. AgriPotential is the first public dataset designed specifically for agricultural potential prediction, aiming to improve data-driven approaches to sustainable land use planning. The dataset and the code are freely accessible at: https://zenodo.org/records/15551829

Three-dimensional brain tumor MRI segmentation is a challenging task in the field of medical image analysis. Recently, deep learning methods, particularly CNN-based architectures, have significantly improved segmentation by learning complex spatial features. However, challenges such as preserving fine details, texture variation, handling class imbalance, and ensuring generalization persist. Enhancing deep learning models with domain-specific knowledge, such as texture-aware filters, can further improve segmentation accuracy and robustness. This paper presents a novel model that incorporates Gabor convolution into a U-Net architecture to enhance texture analysis and minimize feature loss. The model processes 3D brain tumor MRIs slice by slice, utilizing multi-view inputs to preserve spatial details while maintaining a lightweight design. Furthermore, it investigates the optimal kernel sizes for the Gabor filter, marking the first study to address this crucial aspect of integrating textural analysis with deep learning techniques. Experimental results show that the proposed framework improved segmentation accuracy by using a 7×7 Gabor kernel size and achieving Dice coefficients of 89.78% for WT, 85.60% for TC, and 83.55% for ET, with a mean Dice score of 86.31%. The proposed model demonstrated consistent improvements over the standard U-Net and outperformed several existing state-of-the-art methods.

Attention Deficit Hyperactivity Disorder is a neurodevelopmental disorder whose manifestations vary significantly depending on the context. This situational variability poses major challenges for assessing and understanding symptoms, particularly outside clinical environments. In this work, we propose a framework that integrates a contextual vision to enrich medical information. The framework is composed of three main components: a modular ontology that formalizes both medical and contextual dimensions of ADHD; a multi-agent system powered by large language models for automatically extracting and populating knowledge from heterogeneous data sources; and a clinical rule-based reasoning mechanism capable of inferring high-level interpretations from instantiated data. Experimental results demonstrate the framework’s ability to generate accurate, context-sensitive interpretations of symptom manifestations. This approach lays the groundwork for more personalized, explainable, and context-aware patient monitoring, with promising applications in intelligent healthcare systems.

In the literature on deep neural networks, there is considerable interest in developing activation functions that can enhance neural network performance. In recent years, there has been renewed scientific interest in proposing activation functions that can be trained throughout the learning process, as they appear to improve network performance, especially by reducing overfitting. In this paper, we propose a trainable activation function whose parameters need to be estimated. A fully Bayesian model is developed to automatically estimate from the learning data both the model weights and activation function parameters. An MCMC-based optimization scheme is developed to build the inference. The proposed method aims to solve the aforementioned problems and improve convergence time by using an efficient sampling scheme that guarantees convergence to the global maximum. The proposed scheme has been tested across a diverse datasets, encompassing both classification and regression tasks, and implemented in various CNN architectures to demonstrate its versatility and effectiveness. Promising results demonstrate the usefulness of our proposed approach in improving models accuracy due to the proposed activation function and Bayesian estimation of the parameters.

Wearable devices dotted with accelerometers are widely used to study motor overflow in children with attention deficit/hyperactivity disorder (ADHD), as they provide accurate and reliable measurements of physical activity. To accurately determine the time spent in different intensities of physical activity (sedentary, light, moderate or vigorous), it is necessary to identify periods when devices are worn or not. However, this can be problematic because children’s sedentary activities may be mistaken for periods when devices are not worn.In this paper we propose a machine learning approach to detect non-wear periods using data collected from 18 children with ADHD and 18 healthy children using the triaxial accelerometer Actigraph GT9X worn on the non-dominant wrist for one week. The objective is to reduce the overestimation of time spent in activities obtained after data reduction with algorithms using long non-wear time detection periods.The agreement between real data and the classification performed by SVM model (Concordance Correlation Coefficient > 0.95) supported the use of reduced time intervals for detecting non-wear periods.

To assess physical activity intensity using raw acceleration data, various thresholds have been proposed based on different metrics, making it challenging to select the appropriate threshold, particularly when trying to find a threshold adapted to a similar study in terms of device type, population, and device placement. In this study, a cross-validation method is proposed that does not take into account the specific details of the data recording device, the placement of the device, and the characteristics of the population being recorded. We collected acceleration data from 18 healthy children and 18 children with attention deficit/hyperactivity disorder (ADHD) in normal living conditions using the Actigraph GT9X. After collection, the raw data underwent processing steps such as preprocessing, segmentation, and metric extraction. Subsequently, five intensity thresholds were applied to this data, and a voting method using an aggregation approach was used to combine the classifications from each threshold to obtain a final classification. The findings indicated that 97.2% of the voting classifications are reliable (total and approximate decisions) for children with ADHD (97.4% for healthy children), while 2.8% (2.6% for healthy children) should be considered with caution. In conclusion, our approach is flexible and adaptable to different devices and population groups, making it a valuable tool for assessing physical activity in various research contexts.

Mild cognitive impairment (MCI) is a condition marked by impairment in one or more cognitive areas, but not necessarily all of them. It is frequently referred to as the stage between typical age-related cognitive decline and dementia. Recent studies had focused on different modalities to assess disorders such as dementia and Alzheimer's disease (AD). Heart rate variability (HRV) stands out among them as having the potential to identify MCI. In this paper, we propose a new MCI detection method using HRV signals. MCI patients were compared to age-matched healthy controls (HC) for the effect of performing additional cognitive and postural tasks. Twenty-four participants were enrolled to complete three tasks: a postural balance master task, two cognitive tasks called CERAD+ and Neurotrack, and baseline. HRV data were recorded during these experiments. Six machine learning (ML) models were examined for task classification including k-Nearest Neighbors, Decision tree, Random Forest, Extra Trees, Gradient Boosting, and XGBoost. To avoid over-fitting, cross-validation (CV) was employed to assess how well the built models performed. To boost accuracy, a voting ensemble classifier model is developed that combines the top ML models with the highest accuracy rates. The findings of this study demonstrated that MCI might be diagnosed with ML classifiers utilizing HRV signals, particularly when postural and cognitive functions are taken into account.

There is a significant focus in deep neural networks on finding activation functions that can enhance network performance. Scientists are also interested in developing flexible activation functions that can adapt during training to avoid overfitting and provide reliable parameter learning.This paper presents a Bayesian algorithm that employs Markov Chain Monte Carlo (MCMC) to estimate activation function parameters and model weights. This algorithm accelerates convergence through effective sampling and evaluates its effectiveness on two datasets, achieving high accuracy with low complexity.

Hypo-vigilance detection is becoming an important active research areas in the biomedical signal processing field. For this purpose, electroencephalogram (EEG) is one of the most common modalities in drowsiness and awakeness detection. In this context, we propose a new EEG classification method for detecting fatigue state. Our method makes use of a and awakeness detection. In this context, we propose a new EEG classification method for detecting fatigue state. Our method makes use of a Convolutional Neural Network (CNN) architecture. We define an experimental protocol using the Emotiv EPOC+ headset. After that, we evaluate our proposed method on a recorded and annotated dataset. The reported results demonstrate high detection accuracy (93%) and indicate that the proposed method is an efficient alternative for hypo-vigilance detection as compared with other methods.

Using Reflectance Confocal Microscopy (RCM) for lentigo diagnosis is today considered essential. Indeed, RCM allows fast data acquisition with a high spatial resolution of the skin. In this paper, we use a deep convolutional neural network (CNN) to perform RCM image classification in order to detect lentigo. The proposed method relies on an InceptionV3 architecture combined with data augmentation and transfer learning. The method is validated on RCM data and shows very efficient detection performance with more than 98% of accuracy.

Sparse signal/image recovery is a challenging topic that has captured a great interest during the last decades. To address the ill-posedness of the related inverse problem, regularization is often essential by using appropriate priors that promote the sparsity of the target signal/image. In this context, ℓ0+ ℓ1regularization has been widely investigated. In this paper, we introduce a new prior accounting simultaneously for both sparsity and smoothness of restored signals. We use a Bernoulli-generalized Gauss-Laplace distribution to perform ℓ0+ ℓ1+ ℓ2regularization in a Bayesian framework. Our results show the potential of the proposed approach especially in restoring the non-zero coefficients of the signal/image of interest.

In standard within-subject analyses of event-related functional magnetic resonance imaging (fMRI) data, two steps are usually performed separately: detection of brain activity and estimation of the hemodynamic response. Because these two steps are inherently linked, we adopt the so-called region-based joint detection-estimation (JDE) framework that addresses this joint issue using a multivariate inference for detection and estimation. JDE is built by making use of a regional bilinear generative model of the BOLD response and constraining the parameter estimation by physiological priors using temporal and spatial information in a Markovian model. In contrast to previous works that use Markov Chain Monte Carlo (MCMC) techniques to sample the resulting intractable posterior distribution, we recast the JDE into a missing data framework and derive a variational expectation-maximization (VEM) algorithm for its inference. A variational approximation is used to approximate the Markovian model in the unsupervised spatially adaptive JDE inference, which allows automatic fine-tuning of spatial regularization parameters. It provides a new algorithm that exhibits interesting properties in terms of estimation error and computational cost compared to the previously used MCMC-based approach. Experiments on artificial and real data show that VEM-JDE is robust to model misspecification and provides computational gain while maintaining good performance in terms of activation detection and hemodynamic shape recovery.

We address the issue of jointly detecting brain activity and estimating brain hemodynamics from functional MRI data. To this end, we adopt the so-called Joint-Detection-Estimation (JDE) framework introduced in [1] and augmented in [2]. An inherent difficulty is to find the right spatial scale at which brain hemodynamics estimation makes sense. The voxel level is clearly not appropriate as estimating a full hemodynamic response function (HRF) from a single voxel time course may suffer from a poor signal-to-noise-ratio and lead to potentially misleading results (non-physiological HRF shapes). More robust estimation can be obtained by considering groups of voxels (i.e. parcels) with some functional homogeneity properties. Current JDE approaches are therefore based on an initial parcellation but with no guarantee of its optimality or goodness. In this work, we propose a joint parcellation-detection-estimation (JPDE) procedure that incorporates an additional parcel estimation step solving this way both the parcellation choice and robust HRF estimation issues. As in [3], inference is carried out in a Bayesian setting using variational approximation techniques for computational efficiency.

Parallel magnetic resonance imaging (pMRI) relying on multiple receiver coils has emerged as a powerful 3D imaging technique for reducing scanning time or increasing spatial or temporal resolution. The acquired k-space is subsampled, and full field of view (FoV) images are then reconstructed from the acquired aliased data by applying methods such as the SENSE algorithm. However, reconstructed images using SENSE may suffer from several kinds of artifacts mainly because of noise and inaccurate sensitivity profiles. In this paper, we propose a regularized SENSE reconstruction method in which the regularization takes place in the wavelet transform domain. More precisely, a Bayesian strategy is adopted by introducing a bivariate prior to model the complex-valued signal. Experiments on synthetic data and real T1-weighted MRI images at 1.5 Tesla magnetic field show that the proposed method provides improved reconstruction.