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.