Andrea Zanola

Objective: Eating disorders are highly heterogeneous, challenging diagnostic classification and treatment personalisation. This study aimed to identify data‐driven phenotypes of anorexia nervosa and bulimia nervosa and examine their relationship with DSM‐5 diagnoses and short‐term treatment outcomes. Methods: We analysed 809 female patients with anorexia nervosa (restricting or binge–purge subtype) or bulimia nervosa treated in a specialist eating disorders unit. Baseline clinical, sociodemographic, eating‐disorder–specific, and general psychopathology variables were assessed. Nineteen variables entered a Repeated Spectral Clustering pipeline, while body mass index and related measures were reserved for external validation. Associations with DSM‐5 diagnoses and remission at discharge were examined. Results: A four‐cluster solution best fit the data. Two clusters largely corresponded to restricting‐type anorexia nervosa and bulimia nervosa, respectively. Two additional diagnostically mixed clusters were characterised by either greater psychopathological severity and trauma exposure or by older age, longer illness duration, and more marked malnutrition. Cluster membership was associated with remission at the end of treatment, though the effect was weak and no pairwise contrast survived correction. Descriptively, full remission was most frequent in the bulimia‐like cluster. Conclusions: Computational phenotyping captured clinically meaningful heterogeneity beyond DSM‐5 categories including two diagnostically mixed subgroups not identified by categorical diagnosis. Cluster membership at intake was weakly associated with end‐of‐treatment remission, and the resulting subgroups may provide candidate strata for precision psychiatry approaches in eating disorders.

Post-stroke rehabilitation is a complex process influenced by several neurophysiological factors. The recovery is traditionally predicted based on initial

Electroencephalography (EEG) is an important noninvasive diagnostic tool for detecting neurodegenerative disorders. In this context, EEG-based deep learning models show promise due to their ability to capture nonlinear brain dynamics but often suffer from poor generalizability caused by high inter-subject variability. Self-supervised learning (SSL) offers a promising solution by enabling models to learn robust representations from large unlabeled datasets. This study introduces a double-masking representation learning framework for EEG analysis. Using data aggregated from eight multi-center datasets (3156 subjects; 439 h of EEG recordings), a hybrid convolutional-transformer model (TransformEEG) is pretrained to enhance generalization in neurodegenerative disease classification, focusing on Parkinson’s and Alzheimer’s diseases. This approach combines phase-swap data augmentation, designed to facilitate the learning of EEG phase-amplitude coupling, with a double-masking function that operates at both the signal and the transformer’s token levels. The pretrained model was evaluated against a fully supervised baseline using Monte Carlo cross-validation with 100 splits across three public pathology detection datasets. Pretraining led to consistent improvements in both median balanced accuracy and Inter-Quartile Range (IQR) across all fine-tuning datasets. Compared to the fully supervised baseline, the proposed approach increases median balanced accuracy by 2.8% for Parkinson’s disease detection and by 4.2% for Alzheimer’s disease detection, while also reducing performance variability across 100 Monte Carlo splits. These results demonstrate that SSL can enhance EEG deep learning performance, though achieving robust generalization in clinical applications remains an open research challenge.

Psychiatric disorders, including eating disorders (EDs), are characterized by heterogeneity that limits diagnostic accuracy and treatment personalization. Precision psychiatry calls for tools to identify data-driven phenotypes beyond traditional categories. In a large cohort (N=809), we applied an unsupervised clustering pipeline to self-report and clinical variables to uncover ED subgroups. Repeated Spectral Clustering revealed four phenotypes: two aligned with DSM-5 diagnoses (anorexia nervosa restricting-type; bulimia nervosa), and two diagnostically mixed clusters, one characterized by higher psychopathology and trauma exposure, the other by prolonged illness duration. These clusters showed distinct profiles and outcomes. The higher predictability of data-driven clusters compared to DSM-5 categories suggests that unsupervised stratification may offer a complementary perspective on relevant heterogeneity. Our findings highlight how computational phenotyping can advance precision psychiatry by revealing meaningful patterns, informing prognosis, and guiding personalized interventions. This approach may help bridge the gap between clinical presentations and the need for stratified, patient-centered care.

Objective. This work presents xEEGNet, a novel, compact, and explainable neural network for electroencephalography (EEG) data analysis. It is fully interpretable and reduces overfitting through a major parameter reduction. Approach. As an applicative use case to develop our model, we focused on the classification of common dementia conditions, Alzheimer’s and frontotemporal dementia, versus controls. xEEGNet, however, is broadly applicable to other neurological conditions involving spectral alterations. We used ShallowNet, a simple and popular model in the EEGNet family, as a starting point. Its structure was analyzed and gradually modified to move from a ‘black box’ model to a more transparent one, without compromising performance. The learned kernels and weights were analyzed from a clinical standpoint to assess their medical significance. Model variants, including ShallowNet and the final xEEGNet, were evaluated using a robust nested-leave-n-subjects out cross-validation for unbiased performance estimates. Variability across data splits was explained using the embedded EEG representations, grouped by class and set, with pairwise separability to quantify group distinction. Overfitting was measured through training-validation loss correlation and training speed. Main results. xEEGNet uses only 168 parameters, 200 times fewer than ShallowNet, yet retains interpretability, resists overfitting, achieves comparable median performance (−1.5%), and reduces performance variability across splits. This variability is explained by the embedded EEG representations: higher accuracy correlates with greater separation between test-set controls and Alzheimer’s cases, without significant influence from the training data. Significance. The capability of xEEGNet to filter specific EEG bands, learns band specific topographies and use the right EEG spectral bands for disease classification demonstrates its interpretability. While big deep learning models are typically prioritized for performance, this study shows that smaller architectures like xEEGNet can be equally effective in pathology classification, using EEG data.

Electroencephalography (EEG) is establishing itself as an important, low-cost, noninvasive diagnostic tool for the early detection of Parkinson's Disease (PD). In this context, EEG-based Deep Learning (DL) models have shown promising results due to their ability to discover highly nonlinear patterns within the signal. However, current state-of-the-art DL models suffer from poor generalizability caused by high inter-subject variability. This high variability underscores the need for enhancing model generalizability by developing new architectures better tailored to EEG data. This paper introduces TransformEEG, a hybrid Convolutional-Transformer designed for Parkinson's disease detection using EEG data. Unlike transformer models based on the EEGNet structure, TransformEEG incorporates a depthwise convolutional tokenizer. This tokenizer is specialized in generating tokens composed by channel-specific features, which enables more effective feature mixing within the self-attention layers of the transformer encoder. To evaluate the proposed model, four public datasets comprising 290 subjects (140 PD patients, 150 healthy controls) were harmonized and aggregated. A 10-outer, 10-inner Nested-Leave-N-Subjects-Out (N-LNSO) cross-validation was performed to provide an unbiased comparison against seven other consolidated EEG deep learning models. TransformEEG achieved the highest balanced accuracy's median (78.45%) as well as the lowest interquartile range (6.37%) across all the N-LNSO partitions. When combined with data augmentation and threshold correction, median accuracy increased to 80.10%, with an interquartile range of 5.74%. In conclusion, TransformEEG produces more consistent and less skewed results. It demonstrates a substantial reduction in variability and more reliable PD detection using EEG data compared to the other investigated models.

Post-stroke rehabilitation is a complex process influenced by several neurophysio-logical factors. The recovery is traditionally predicted based on initial impairment using linear models. The Proportional Recovery Rule (PRR), developed on the Fugl-Meyer scale, has even been proposed as a therapeutic target. In this framework, patients are classified as “fitters” or “non-fitters”, though this distinction depends on the methodology used. Additionally, issues like mathematical coupling and ceiling effects on clinical scales could raise concerns about the validity of these models. To overcome these issues, Repeated Spectral Clustering (RSC) was used to identify recovery patterns based on NIHSS. We selected 201 patients from the WAKE-UP trail, all moderately impaired at onset and still impaired at 22/36 hours. Clustering was performed using a similarity matrix based on pairwise absolute differences between recovery ratios, calculated from 22/36 hours to 90 days post-stroke. Cluster differences were tested with prognostic factors, including lesion volume, side, treatment, and the Heidelberg scale. The PRR was fit to the cohort for comparison with clustering results. The linear fit reproduced findings consistent with the literature, such as a correlation of ρ ( x , Δ) = . 73 and an average recovery ratio of 70% for the “fitters”. RSC grouped patients into six recovery clusters: C ₀ (full recovery), C ₁ (above average), C ₂ and C ₃ (average, PRR-aligned), C ₄ (below average), and C ₅ (deterioration). NIHSS scores in most patients declined non-proportionally. Lesion volume was not significantly different across clusters, while left-sided strokes were higher in low recovery clusters. Patients with a recovery ratio ≥ 0.3 within two weeks mostly fell into favorable clusters ( C ₀ – C ₃ ), covering ≈ 90% of such cases. The identified clusters provide a refined view of stroke recovery following wake-up stroke. Clustering better captures patient similarities, enabling the assessment of neurophysiological differences between groups and supporting tailored interventions.

Deep learning is significantly advancing the analysis of electroencephalography (EEG) data by effectively discovering highly nonlinear patterns within the signals. Data partitioning and cross-validation are crucial for assessing model performance and ensuring study comparability, as they can produce varied results and data leakage due to specific signal properties (e.g., biometric). Such variability leads to incomparable studies and, increasingly, overestimated performance claims, which are detrimental to the field. Nevertheless, no comprehensive guidelines for proper data partitioning and cross-validation exist in the domain, nor is there a quantitative evaluation of their impact on model accuracy, reliability, and generalizability. To assist researchers in identifying optimal experimental strategies, this paper thoroughly investigates the role of data partitioning and cross-validation in evaluating EEG deep learning models. Five cross-validation settings are compared across three supervised cross-subject classification tasks (BCI, Parkinson's, and Alzheimer's disease detection) and four established architectures of increasing complexity (ShallowConvNet, EEGNet, DeepConvNet, and Temporal-based ResNet). The comparison of over 100,000 trained models underscores, first, the importance of using subject-based cross-validation strategies for evaluating EEG deep learning models, except when within-subject analyses are acceptable (e.g., BCI). Second, it highlights the greater reliability of nested approaches (N-LNSO) compared to non-nested counterparts, which are prone to data leakage and favor larger models overfitting to validation data. In conclusion, this work provides EEG deep learning researchers with an analysis of data partitioning and cross-validation and offers guidelines to avoid data leakage, currently undermining the domain with potentially overestimated performance claims.