publications

2026

1. Multimodal MRI neurodevelopmental profiling in type 1 diabetes: long-term effects of MDI vs CSII treatments

   Pasquale Anthony Della Rosa, Nicolò Pecco, Sonia Francesca Calloni, and 16 more authors

   The Journal of Clinical Endocrinology & Metabolism, Mar 2026

   Abs DOI

   The developing brain is particularly vulnerable to glycemic extremes in early-onset type 1 diabetes (T1D). However, how treatment-specific modalities may influence long-term neurodevelopmental trajectories remains poorly understood.To characterize multimodal MRI neurodevelopmental profiles in pediatric T1D and evaluate treatment-related effects of multiple daily injections (MDI) vs continuous subcutaneous insulin infusion (CSII) on brain structure, function, and test whether glycated hemoglobin (HbA1c)-linked imaging features relate to executive–working-memory performance.Sixteen children with T1D (8 MDI and 8 CSII from diagnosis) and eight controls underwent structural MRI, diffusion MRI, and resting-state fMRI. Union Recursive Feature Elimination selected gray matter (GM), white matter (WM), and resting-state functional connectivity (rs-FC) features discriminating groups; regression related selected features to long-term age-adjusted mean HbA1c. NEPSY-II Word List Interference (WI) was administered; control-referenced WI outcomes were examined vs HbA1c and HbA1c-associated structural features, including mediation.Functional features outperformed structural features (balanced accuracy 0.83 vs 0.67). MDI showed reduced GM/WM integrity and disrupted fronto-temporal and subcortical connectivity vs CSII and controls. Right inferior frontal gyrus (IFG) volume correlated with HbA1c (r = 0.71, P < .05) and predicted HbA1c (β = 0.28, P = .015). Higher HbA1c related to poorer WI repetition (r = −0.60, P = .013), and right IFG volume related to poorer WI repetition (r = −0.70, P = .002). Mediation supported an indirect HbA1c effect via right IFG volume (a × b = −0.676; Sobel z = −1.765, one-tailed P = .0388), explaining ∼64% of the total association. CSII had 30% lower hyperglycemia exposure than MDI and higher WI repetition mean ranks (11.19 vs 5.81; P = .023).Pediatric T1D is associated with multimodal neuroimaging alterations influenced by insulin treatment modality. CSII may confer neuroprotective benefits by improving metabolic control and preserving functional connectivity. Right IFG volume is a candidate imaging marker linking metabolic regulation to interference-sensitive executive–working-memory vulnerability.

2. Do You Understand How I Feel?: Towards Verified Empathy in Therapy Chatbots

   Francesco Dettori, Matteo Forasassi, Lorenzo Veronese, and 3 more authors

   Mar 2026

   Abs DOI Bib PDF Code

   Conversational agents are increasingly used as support tools along mental therapeutic pathways with significant societal impacts. In particular, empathy is a key non-functional requirement in therapeutic contexts, yet current chatbot development practices provide no systematic means to specify or verify it. This paper envisions a framework integrating natural language processing and formal verification to deliver empathetic therapy chatbots. A Transformer-based model extracts dialogue features, which are then translated into a Stochastic Hybrid Automaton model of dyadic therapy sessions. Empathy-related properties can then be verified through Statistical Model Checking, while strategy synthesis provides guidance for shaping agent behavior. Preliminary results show that the formal model captures therapy dynamics with good fidelity and that ad-hoc strategies improve the probability of satisfying empathy requirements.

   @misc{DettoriForasassiVeronese2026,
     title = {Do You Understand How I Feel?: Towards Verified Empathy in Therapy Chatbots},
     author = {Dettori, Francesco and Forasassi, Matteo and Veronese, Lorenzo and Lestingi, Livia and Scotti, Vincenzo and Rossi, Matteo Giovanni},
     year = {2026},
     eprint = {2601.08477},
     doi = {10.48550/arXiv.2601.08477},
     archiveprefix = {arXiv},
     primaryclass = {cs.CL},
     url = {https://arxiv.org/abs/2601.08477},
   }

2025

1. Optimized AI-based neural decoding from BOLD fMRI signal for analyzing visual and semantic ROIs in the human visual system

   Lorenzo Veronese, Andrea Moglia, Nicolò Pecco, and 4 more authors

   Journal of Neural Engineering, Aug 2025

   Abs DOI Bib PDF Code Website

   Objective. AI-based neural decoding reconstructs visual perception by leveraging generative models to map brain activity measured through functional magnetic resonance imaging (fMRI) into the observed visual stimulus. Approach. Traditionally, ridge linear models transform fMRI into a latent space, which is then decoded using variational autoencoders (VAE) or LDMs. Owing to the complexity and noisiness of fMRI data, newer approaches split the reconstruction into two sequential stages, the first one providing a rough visual approximation using a VAE, the second one incorporating semantic information through the adoption of LDM guided by contrastive language-image pre-training (CLIP) embeddings. This work addressed some key scientific and technical gaps of the two-stage neural decoding by: (1) implementing a gated recurrent unit-based architecture to establish a non-linear mapping between the fMRI signal and the VAE latent space, (2) optimizing the dimensionality of the VAE latent space, (3) systematically evaluating the contribution of the first reconstruction stage, and (4) analyzing the impact of different brain regions of interest (ROIs) on reconstruction quality. Main results. Experiments on the NSD, containing 73 000 unique natural images, along with fMRI of eight subjects, demonstrated that the proposed architecture maintained competitive performance while reducing the complexity of its first stage by 85%. The sensitivity analysis showcased that the first reconstruction stage is essential for preserving high structural similarity in the final reconstructions. Restricting analysis to semantic ROIs, while excluding early visual areas, diminished visual coherence, preserving semantics though. The inter-subject repeatability across ROIs was about 92% and 98% for visual and sematic metrics, respectively. Significance. This study represents a key step toward optimized neural decoding architectures leveraging non-linear models for stimulus prediction. Sensitivity analysis highlighted the interplay between the two reconstruction stages, while ROI-based analysis provided strong evidence that the two-stage AI model reflects the brain’s hierarchical processing of visual information.

   @article{Veronese2025,
     author = {Veronese, Lorenzo and Moglia, Andrea and Pecco, Nicolò and Anthony Della Rosa, Pasquale and Scifo, Paola and Mainardi, Luca and Cerveri, Pietro},
     journal = {Journal of Neural Engineering},
     title = {Optimized AI-based neural decoding from BOLD fMRI signal for analyzing visual and semantic ROIs in the human visual system},
     year = {2025},
     issn = {1741-2552},
     month = aug,
     number = {4},
     pages = {046048},
     volume = {22},
     doi = {10.1088/1741-2552/adfbc2},
     publisher = {IOP Publishing},
   }

2023

1. Stain Transfer using CycleGAN for Histopathological Images

   Lorenzo Veronese, Isabella Poles, Eleonora D’Arnese, and 1 more author

   In IEEE EUROCON 2023 - 20th International Conference on Smart Technologies, Aug 2023

   Abs DOI Bib PDF

   Histopathology refers to the observation of tissues to identify the manifestation of diseases, e.g., cancer. Tiny tissue samples are taken from the patient and studied through a microscope; the analysis of the different cells, particularly their nuclei and other structures, allows for disease detection. The biological specimens need some preparation, namely Hematoxylin and Eosin (H&E) staining is often used to highlight nuclei and cytoplasm. Although staining is fundamental, given that cells are transparent when imaged, it is still highly affected by casual errors: colors change when a small preparation step is slightly different and even when a different microscope is used. This factor leads to Computer Aided Detection (CAD) systems losing performance. Therefore, to solve this problem and allow for the integration of multiple low-dimensional datasets, we propose a CycleGAN-based architecture exploiting PatchGAN and U-Net backbones as discriminators and generators, respectively, demonstrating an improvement in mean Structural Similarity Index Measure (SSIM) over the one computed on the original datasets of around 1.8%.

   @inproceedings{Veronese2023,
     author = {Veronese, Lorenzo and Poles, Isabella and D'Arnese, Eleonora and Santambrogio, Marco D.},
     booktitle = {IEEE EUROCON 2023 - 20th International Conference on Smart Technologies},
     title = {Stain Transfer using CycleGAN for Histopathological Images},
     year = {2023},
     volume = {},
     number = {},
     pages = {1-5},
     keywords = {Design automation;Image color analysis;Histopathology;Microscopy;Generators;Biology;Indexes},
     doi = {10.1109/EUROCON56442.2023.10199027},
   }