Autonomous navigation systems for Mars rovers face significant challenges due to the limited availability of real Martian imagery for training Artificial Intelligence (AI) models. In contrast, a large amount of synthetic images of Martian landscape can be generated and used for training. This paper presents a novel "Sim2Real" approach that integrates synthetic data generation and classification with few-shot learning to enhance real images classification for autonomous Mars rovers. Specifically, a custom Convolutional Neural Network (CNN) is first developed and pre-trained on synthetic Martian landscapes provided by Thales Alenia Space Italy (TAS-I). The pre-trained CNN is then used as a basis to develop the proposed Sim2Real approach. In particular, a few-shot learning strategy is introduced to classify real Mars rover images from an existing available dataset of the National Aeronautics and Space Administration (NASA). Preliminary experimental results demonstrated promising performance when applied to real Martian images, highlighting the potential of few-shot based Sim2Real approaches for planetary exploration.

This paper introduces a hybrid deep learning approach for processing electroencephalographic (EEG) signals to track their dynamics over time in the spatial-spectral domain. This method is particularly valuable whenever the temporal evolution of the brain process under analysis is relevant to classification. A hybrid deep learning model, herein referred to as EEGConViT, processes spatial-temporal stream of data (EEG movies). Specifically, EEGConViT consists of a custom CNN that encodes each frame into a feature vector, which is then augmented with temporal position embeddings with a Transformer model able to capture sequential dependencies. In this work, the application of motor EEG signals was investigated with the proposed EEGConViT. In particular, EEG signals preceding motor execution were processed to assess the ability of the model to predict the upcoming sub-movement of the upper limb. The model was evaluated using a collection of EEG signals from 14 subjects, derived from a publicly available repository. Using a leave-one-subject-out strategy, the model was trained on data from 13 subjects and fine-tuned on the remaining one (cross-subject training and subsequent calibration over the single subject). Results demonstrate that our approach outperforms comparable models in the literature while significantly reducing training time, an essential factor in medical applications, where both classification performance and rapid calibration are critical.

Decoding electroencephalographic (EEG) signals is of key importance in the development of brain–computer interface (BCI) systems. However, high inter-subject variability in EEG signals requires user-specific calibration, which can be time-consuming and limit the application of deep learning approaches, due to general need of large amount of data to properly train these models. In this context, this paper proposes a multidimensional and explainable deep learning framework for fast and interpretable EEG decoding. In particular, EEG signals are projected into the spatial–spectral–temporal domain and processed using a custom three-dimensional (3D) Convolutional Neural Network, here referred to as EEGCubeNet. In this work, the method has been validated on EEGs recorded during motor BCI experiments. Namely, hand open (HO) and hand close (HC) movement planning was investigated by discriminating them from the absence of movement preparation (resting state, RE). The proposed method is based on a global- to subject-specific fine-tuning. The model is globally trained on a population of subjects and then fine-tuned on the final user, significantly reducing adaptation time. Experimental results demonstrate that EEGCubeNet achieves state-of-the-art performance (accuracy of [Formula: see text] and [Formula: see text] for HC versus RE and HO versus RE, binary classification tasks, respectively) with reduced framework complexity and with a reduced training time. In addition, to enhance transparency, a 3D occlusion sensitivity analysis-based explainability method (here named 3D xAI-OSA) that generates relevance maps revealing the most significant features to each prediction, was introduced. The data and source code are available at the following link: https://github.com/AI-Lab-UniRC/EEGCubeNet

Skin cancer is one of the leading causes of mortality worldwide. Early diagnosis can ensure more effective patient treatment and outcomes, but, this is challenging due to the high similarity between different skin lesion types. There is a growing interest in developing Artificial Intelligence (AI)-based systems for automated skin lesion classification. However, current AI models are not transparent, leading to a lack of trust from clinicians who struggle to interpret and validate AI decisions. To this end, in this paper, a fine tuned EfficientNet-B0-based classifier is first developed to classify dermoscopic images of Melanoma (MEL), Nevus (NV) and Seborrheic Keratosis (SK) skin lesions gathered from the International Skin Imaging Collaboration (ISIC) dataset. Next, the explainability of the model is investigated. In particular, a new Trustworthiness Index for eXplainable AI, herein referred to as TIxAI , is proposed. The TIxAI is based on the difference between the relevance degree of the lesion and non-lesion areas, leading to the conclusion that the higher the TIxAI , the more trustworthy the classifier is expected to be. Experimental results support the use of the proposed TIxAI to assess and benchmark the reliability of classifiers also in other real-world applications.

Electroencephalography (EEG) is widely used as a non-invasive technique for the diagnosis of several brain disorders, including Alzheimer's disease and epilepsy. Until recently, diseases have been identified over EEG readings by human experts, which may not only be specific and difficult to find, but are also subject to human error. Despite the recent emergence of machine learning methods for the interpretation of EEGs, most approaches are not capable of capturing the underlying arbitrary non-Euclidean relations between signals in the different regions of the human brain. In this context, Graph Neural Networks (GNNs) have gained attention for their ability to effectively analyze complex relationships within different types of graph-structured data. This includes EEGs, a use case still relatively unexplored. In this paper, we aim to bridge this gap by presenting a study that applies GNNs for the EEG-based detection of Alzheimer's disease and discrimination of two different types of seizures. To this end, we demonstrate the value of GNNs by showing that a single GNN architecture can achieve state-of-the-art performance in both use cases. Through design space explorations and explainability analysis, we develop a graph-based transformer that achieves cross-validated accuracies over 89% and 96% in the ternary classification variants of Alzheimer's disease and epilepsy use cases, respectively, matching the intuitions drawn by expert neurologists. We also argue about the computational efficiency, generalizability and potential for real-time operation of GNNs for EEGs, positioning them as a valuable tool for classifying various neurological pathologies and opening up new prospects for research and clinical practice.

Automatic ship detection from spaceborne systems such as satellites or aircrafts, raises considerable attention in sea surface monitoring because of the several applications in military and civilian field. In this context, processing satellite images on-board would reduce the latency time especially for emergency situations. In this paper, an hardware-oriented (HO) ship detection system based on a customized Convolutional Neural Network (CNN), here referred to as HO-ShipNet, is proposed and tested on a revised version of the “Ships in Satellite Imagery” (SSI) Kaggle dataset, reporting detection accuracy of up to 95%. Furthermore, the explainability of HO-ShipNet is investigated by means of explainable Artificial Intelligence (xAI) techniques (i.e., Local Interpretable Model-Agnostic Explanation (LIME) and Occlusion Sensitivuty Analysis (OSA)), in order to understand the reasoning behind the HO-ShipNet decisions by detecting the most important input features and consequently ensure the trustworthiness of the model itself. Finally, HO-ShipNet is also implemented on the heterogeneous Xilinx xc7z045ffg900-2 SoC Field Programmable Gate Array (FPGA) outperforming state-of-the-art FPGA-based accelerators dealing with high-resolution frames. The promising results encourage the potential deployment of the proposed system for on-board applications.

In this study, a few-shot transfer learning approach was introduced to decode movement intention from electroencephalographic (EEG) signals, allowing to recognize new tasks with minimal adaptation. To this end, a dataset of EEG signals recorded during the preparation of complex sub-movements was created from a publicly available data collection. The dataset was divided into two parts: the source domain dataset (including 5 classes) and the support (target domain) dataset, (including 2 classes) with no overlap between the two datasets in terms of classes. The proposed methodology consists in projecting EEG signals into the space-frequency-time domain, in processing such projections (rearranged in channels × frequency frames) by means of a custom EEG-based deep neural network (denoted as EEGframeNET5), and then adapting the system to recognize new tasks through a few-shot transfer learning approach. The proposed method achieved an average accuracy of 72.45 ± 4.19% in the 5-way classification of samples from the source domain dataset, outperforming comparable studies in the literature. In the second phase of the study, a few-shot transfer learning approach was proposed to adapt the neural system and make it able to recognize new tasks in the support dataset. The results demonstrated the system’s ability to adapt and recognize new tasks with an average accuracy of 80 ± 0.12% in discriminating hand opening/closing preparation and outperforming reported results in the literature. This study suggests the effectiveness of EEG in capturing information related to the motor preparation of complex movements, potentially paving the way for BCI systems based on motion planning decoding. The proposed methodology could be straightforwardly extended to advanced EEG signal processing in other scenarios, such as motor imagery or neural disorder classification.

The present paper introduces a novel method, named AutoEncoder-Filter Bank Common Spatial Patterns (AE-FBCSP), to decode imagined movements from electroencephalography (EEG). AE-FBCSP is an extension of the well-established FBCSP and is based on a global (cross-subject) and subsequent transfer learning subject-specific (intra-subject) approach. A multi-way extension of AE-FBCSP is also introduced in this paper. Features are extracted from high-density EEG (64 electrodes), by means of FBCSP, and used to train a custom AE, in an unsupervised way, to project the features into a compressed latent space. Latent features are used to train a supervised classifier (feed forward neural network) to decode the imagined movement. The proposed method was tested using a public dataset of EEGs collected from 109 subjects. The dataset consists of right-hand, left-hand, both hands, both feet motor imagery and resting EEGs. AE-FBCSP was extensively tested in the 3-way classification (right hand vs left hand vs resting) and also in the 2-way, 4-way and 5-way ones, both in cross- and intra-subject analysis. AE-FBCSP outperformed standard FBCSP in a statistically significant way (p > 0.05) and achieved a subject-specific average accuracy of 89.09% in the 3-way classification. The proposed methodology performed subject-specific classification better than other comparable methods in the literature, applied to the same dataset, also in the 2-way, 4-way and 5-way tasks. One of the most interesting outcomes is that AE-FBCSP remarkably increased the number of subjects that responded with a very high accuracy, which is a fundamental requirement for BCI systems to be applied in practice.

In the Covid-19 era, it is important to have an edge detector for X-ray (XR) images affected by uncertainties with low computational load but with high performance. So, here, a new version of a well-known fuzzy edge detector, in which a new image fuzzification procedure has been formulated, is proposed. The performance were qualitatively/ quantitatively compared with those obtained by Canny’s edge detector (gold standard for this type of problem). In addition, an evolution of the deep fuzzy-neural model named CovNNet, recently proposed by the authors to discriminate chest XR (CXR) images of patients with Covid-19 pneumonia from images of patients with interstitial pneumonias not related to Covid-19 (No-Covid-19), is presented and referred as to Enhanced-CovNNet (ECovNNet). Here, the generalization ability of it is also improved by introducing a regularization based on dropping out some nodes of the network in a random way. ECovNNet processes input CXR images and the corresponding fuzzy CXR images (processed through the proposed enhanced-fuzzy edge detector) and extracts relevant CXR/fuzzy features, subsequently combined in a single array named CXR and fuzzy features vector. The latter is used as input to an Autoencoder-(AE)-based classifier to perform the binary classification: Covid-19 and No-Covid-19, reporting accuracy rate up to 81%. Finally, the work is completed with some interesting physico-mathematical results.

In this study a Deep Learning (DL) based-Brain-Computer Interface (BCI) system able to automatically detect and decode voluntary eye blinks from the analysis of electroen-cephalographic (EEG) signals is proposed for controlling, in principle, an external device by means of ocular movements. To this end, a Convolutional Neural Network (CNN) is developed to classify EEG recordings related to natural (or involuntary) blinks, forced (or voluntary) blinks and baseline (no blinks) category. The proposed system achieved an impressive average classification performance: accuracy rate up to 99.4% +/- 1.3%. However, the core of the present study was to investigate the explainability and interpretability of the proposed CNN with the ultimate aim of explore which segments of the EEG signal is the most relevant in the voluntary/involuntary blink discrimination process. To this end, explainable Artificial Intelligence (xAI) techniques were applied. Specifically, the Gradient-weighted Class Activation Mapping (Grad-CAM) and the Local Interpretable Model Agnostic Explanation (LIME) algorithms were used. xAI allowed us to visually identify the most relevant EEG areas especially for the voluntary and involuntary blink detection. Indeed, limited to the analyzed dataset, for natural blinks, the discriminating region was the interval ranged from the temporal instant the eye was closed till the following instants of the reopening (vice-versa for voluntary blinks). The baseline (no blink), on the other hand, was characterized by a low activation threshold throughout the EEG segment.

This paper proposes a generative model and transfer learning powered system for classification of Scanning Electron Microscope (SEM) images of defective nanofibers (D-NF) and nondefective nanofibers (ND-NF) produced by electrospinning (ES) process. Specifically, a conditional-Generative Adversarial Network (c-GAN) is developed to generate synthetic D-NF/ND-NF SEM images. A transfer learning-oriented strategy is also proposed. First, a Convolutional Neural Network (CNN) is pre-trained on real images. The transfer-learned CNN is trained on synthetic SEM images and validated on real ones, reporting accuracy rate up to 95.31%. The achieved encouraging results endorse the use of the proposed generative model in industrial applications as it could reduce the number of needed laboratory ES experiments that are costly and time consuming.

Being able to decode the subject's intention to move is still a major challenge in the field of Brain Computer Interfaces (BCI). Even more, decoding the intention to perform movements from the motor preparation phase is a still largely unexplored topic, as most of the efforts have been focused so far on motor imagery. The present paper deals with BCIs based on electroencephalography (EEG), the best candidate for future systems meant for widespread use, with the goal of decoding the preparation of hand open/close movement from the EEG recordings of the subject. To this end, a dataset of EEG signals recorded in the 1s frame preceding the onset of movement are extracted from a publicly available database. Epochs are properly pre-filtered between 0.5 and 32 Hz and labelled as pre-hand closing (HC), pre-hand opening (HO) or resting (RE) epochs. A system for motion planning decoding, based on a custom Convolutional Neural Network (CNN) and named “Motion Planning Neural Network” MPnnet, is designed, trained and tested over the constructed dataset, achieving a mean HC-RE and HO-RE accuracy of$90.77 \pm 5.56\%$and of$92.48 \pm 4.3\%$, respectively. MPnnet matched the performance of more complex systems proposed in the past, allowing to skip the inverse problem solution step and showing to be able to self-learn relevant features directly from scalp EEG signals.

In this paper, a hybrid-domain deep learning (DL)-based neural system is proposed to decode hand movement preparation phases from electroencephalographic (EEG) recordings. The system exploits information extracted from the temporal-domain and time-frequency-domain, as part of a hybrid strategy, to discriminate the temporal windows (i.e. EEG epochs) preceding hand sub-movements (open/close) and the resting state. To this end, for each EEG epoch, the associated cortical source signals in the motor cortex and the corresponding time-frequency (TF) maps are estimated via beamforming and Continuous Wavelet Transform (CWT), respectively. Two Convolutional Neural Networks (CNNs) are designed: specifically, the first CNN is trained over a dataset of temporal (T) data (i.e. EEG sources), and is referred to as T-CNN; the second CNN is trained over a dataset of TF data (i.e. TF-maps of EEG sources), and is referred to as TF-CNN. Two sets of features denoted as T-features and TF-features, extracted from T-CNN and TF-CNN, respectively, are concatenated in a single features vector (denoted as TTF-features vector) which is used as input to a standard multi-layer perceptron for classification purposes. Experimental results show a significant performance improvement of our proposed hybrid-domain DL approach as compared to temporal-only and time-frequency-only-based benchmark approaches, achieving an average accuracy of [Formula: see text]%.

The goal of this paper is to develop a Brain Computer Interface (BCI) based on voluntary eye blinks decoding. In particular, the study was focused on the signals generated in the cortex by eye blinking, which can be collected by frontopolar scalp Electroencephalographic (EEG) sensors. Normally, EEG recording systems meant for clinical applications are expensive and cannot be used in large-scale user-friendly applications. Thanks to a prototype made by the STMicroelectronics company, based on an Open Source EEG project, a low-cost EEG recording system was created in this work. The goal is to develop an algorithm that can detect and discriminate between voluntary (forced) and involuntary (natural) blinking so that, in the future, an EEG-based BCI system that is able to control a device through eye movements could be developed, which would be of great use for all people with motor disabilities who can control eye movements. The proposed algorithm is based on a one-dimensional (1D) Convolutional Neural Network (CNN) architecture. Frontopolar EEG signals were collected during the execution of voluntary and spontaneous blinks by four healthy subjects. A dataset of EEG epochs of including blinks was constructed and used to train and validate the proposed CNN. The proposed system allowed to discriminate the blinks performed by the subjects (voluntary vs. involuntary) with an average accuracy of 97.92%.

In this paper, a novel deep learning based approach is proposed for the automatic classification of Electroencephalographic (EEG) signals of subjects diagnosed with the dementia of Alzheimer's disease (AD), Mild Cognitive Impairment (MCI) and Healthy Control (HC). Specifically, a custom Convolutional Neural Network (CNN) is designed to receive as input AD/MCI/HC EEG segments (epochs) of the same temporal width, and perform 2-way classification tasks: AD vs. HC, AD vs. MCI, MCI vs. HC. Our proposed architecture, termed EEG-CNN, is shown to exhibit remarkable abilities to self-learn relevant features directly from the EEG traces, avoiding the need for hand-crafted feature extraction engineering. Comparative experimental results demonstrate the promising performance of EEG-CNN, which is based on an analysis of the EEG time series only, reporting accuracies of 85.78 ± 2.18%, 69.03 ± 1.33%, 85.34 ± 1.86% in AD vs. HC, AD vs. MCI and MCI vs. HC classifications, respectively.

In the last decades, the use of Machine Learning (ML) algorithms have been widely employed to aid clinicians in the difficult diagnosis of neurological disorders, such as Alzheimer's disease (AD). In this context, here, a data-driven ML system for classifying Electroencephalographic (EEG) segments (i.e. epochs) of patients affected by AD, Mild Cognitive Impairment (MCI) and Healthy Control (HC) individuals, is introduced. Specifically, the proposed ML system consists of evaluating the average Time-Frequency Map (aTFM) related to a 19-channels EEG epoch and extracting some statistical coefficients (i.e. mean, standard deviation, skewness, kurtosis and entropy) from the main five conventional EEG sub-bands (or EEG-rhythms: delta, theta, alpha1, alpha2, beta). Afterwards, the time-frequency features vector is fed into an Autoeconder (AE), a Multi-Layer Perceptron (MLP), a Logistic Regression (LR) and a Support Vector Machine (SVM) based classifier to perform the 2-ways EEG epoch-classification tasks: AD vs HC and AD vs MCI. The performances of the proposed approach have been evaluated on a dataset of 189 EEG signals (63 AD, 63 MCI and 63 HC), recorded during an eye-closed resting condition at IRCCS Centro Neurolesi Bonino Pulejo of Messina (Italy). Experimental results reported that the 1-hidden layer MLP (MLP1) outperformed all the other developed learning systems as well as recently proposed state-of-the-art methods, achieving accuracy rate up to 95.76 % ± 0.0045 and 86.84% ± 0.0098 in AD vs HC and AD vs MCI classification, respectively.

Alzheimer's disease (AD) is a neurodegenerative disorder that causes a loss of connections between neurons. The goal of this paper is to construct a complex network model of the brain-electrical activity, using high-density EEG (HD-EEG) recordings, and to compare the network organization in AD, mild cognitive impaired (MCI), and healthy control (CNT) subjects. The HD-EEG of 16 AD, 16 MCI, and 12 CNT was recorded during an eye-closed resting state. The permutation disalignment index (PDI) was used to describe the dissimilarity between EEG signals and to construct the connection matrices of the network model. The three groups were found to have significantly different (p <; 0.001) characteristic path length (λ), average clustering coefficient (CC), and the global efficiency (GE). This is the first time that HD-EEG signals of AD, MCI, and CNT have been compared and that PDI has been used to discriminate between the three groups. Considering the large amount of data originating from HD-EEG acquisition, compared to standard EEG, the aim of this paper is also to assess that compression did not alter the results of the complex network analysis. Compressive sensing was adopted to compress and reconstruct the HD-EEG signals with minimal information loss, achieving an average structural similarity index of 0.954 (AD), 0.957 (MCI), and 0.959 (CNT). When applied to the reconstructed HD-EEG, complex network analysis provided a substantially unaltered performance, compared to the analysis of the original signals: λ, CC, and GE of the three groups were indeed still significantly different (p <; 0.001).

In this paper, a novel electroencephalographic (EEG)-based method is introduced for the quantification of brain-electrical connectivity changes over a longitudinal evaluation of mild cognitive impaired (MCI) subjects. In the proposed method, a dissimilarity matrix is constructed by estimating the coupling strength between every pair of EEG signals, Hierarchical clustering is then applied to group the related electrodes according to the dissimilarity estimated on pairs of EEG recordings. Subsequently, the connectivity density of the electrodes network is calculated. The technique was tested over two different coupling strength descriptors: wavelet coherence (WC) and permutation Jaccard distance (PJD), a novel metric of coupling strength between time series introduced in this paper. Twenty-five MCI patients were enrolled within a follow-up program that consisted of two successive evaluations, at time T0 and at time T1, three months later. At T1, four subjects were diagnosed to have converted to Alzheimer's Disease (AD). When applying the PJD-based method, the converted patients exhibited a significantly increased PJD (p < 0.05), i.e., a reduced overall coupling strength, specifically in delta and θ bands and in the overall range (0.5-32 Hz). In addition, in contrast to stable MCI patients, converted patients exhibited a network density reduction in every subband (delta, θ, alpha, and beta). When WC was used as coupling strength descriptor, the method resulted in a less sensitive and specific outcome. The proposed method, mixing nonlinear analysis to a machine learning approach, appears to provide an objective evaluation of the connectivity density modifications associated to the MCI-AD conversion, just processing noninvasive EEG signals.

In this paper, the need of novel methods to extract diagnostic information from the Electroencephalographic (EEG) recordings of epileptic patients was addressed. A novel method, based on Wavelet Coherence (WC) between EEG signals and Hierarchical Clustering (HC), was proposed to estimate the EEG network connectivity density in Childhood Absence Epilepsy (CAE) patients. The EEG recordings of four patients affected by CAE were partitioned into non overlapping windows and WC was estimated window by window. The behaviour of WC was analysed over the time, for every couple of EEG electrodes. The ictal states (seizures) resulted associated to increased WC levels, thus reflecting an increased synchronization between electrodes during the seizure. A WC-based dissimilarity index was then defined and HC was fed with the dissimilarity indices between every pair of electrodes with the aim of finding possible correlations between changes in electrode clustering and changes in the brain state. For every window under analysis, a dendrogram was constructed, the corresponding set of electrode clusters was determined and the subsequent network density values were calculated. Seizures resulted typically associated to increased network density, reflecting an increased connectivity during the ictal states.

Alzheimer's disease (AD) is a degenerative neurological disorder characterized by a loss of functional connections between different areas of the brain. AD is considered a cortical dementia, thus Electroencephalography (EEG) has been used as a tool for diagnosing AD for the last two decades. Often, the hallmarks of EEG abnormality in AD patients are a shift of the power spectrum to lower frequencies and reduced coherences among cortical regions, however, it is still mostly unknown how these abnormalities evolve together with the disease progression. In this paper we proposed a longitudinal study of the EEG of three AD patients in order to study the disease progression, from the coherence point of view, over the four major EEG sub-bands: delta, theta, alpha and beta. The EEG was recorded at time T0 and then after three months (time T1). We proposed a coherence based hierarchical clustering method to group the electrodes together according to their mutual pairwise coherence, in order to evaluate how the brain connectivity changed along with the disease in the spectral domain. The results provide an in-depth view of the structure of electrode interconnection of every single patient in every sub-band at time T0 and time T1. This study endorsed the commonly shared belief that coherence reduces over time but it revealed that coherence spatial distribution changes in a different way, from patient to patient. The results also showed that a patient-specific brain connectivity analysis is possible and that a personalized analysis of the disease's progression might provide valuable diagnostic information. In the near future, the study will be extended to a larger dataset in order to validate the method statistically.