UNIVERSITY OF ROME TOR VERGATA

Medical Physics,
AI and Neurotechnology

Experimental and computational methods for measuring, modelling and understanding biological and neural systems.

Our laboratories ↗ Research themes ↗

TOR VERGATA UNIVERSITY OF ROME

Department of Biomedicine and Prevention

ABOUT THE GROUP

Quantitative methods across scales.

Our group brings together physicists, mathematicians, engineers and life scientists to investigate biological systems through quantitative measurement and modelling. We develop experimental techniques, mathematical models and learning algorithms that connect molecular interactions and cellular processes with tissue properties, neural networks and whole-system physiology. Across these scales, our work links measurable signals to underlying mechanisms, with attention to uncertainty, reproducibility and the limits of inference.

Our laboratories provide complementary experimental and computational capabilities, while research teams form around shared scientific questions and projects. Imaging physics, neural interfaces, artificial intelligence, neuromorphic computing, signal analysis and molecular measurement connect through common methods and collaborative work. Academic staff, research fellows, postdoctoral researchers and doctoral students contribute distinct expertise to the development and validation of these approaches.

Based in the Department of Biomedicine and Prevention at the University of Rome Tor Vergata, the group combines research with university teaching and research training. Our projects are supported by competitive European and national programmes, charitable research funding and advanced computing allocations, alongside scientific collaborations with academic and industrial partners.

02 / RESEARCH

Research themes

Funded projects and collaborations

Research grants, computing allocations and scientific collaborations.

Funded projects ↗   Industrial collaborations ↗

Researchers and staff

Researchers in physics, mathematics, engineering and the life sciences, with expertise in experimental and computational biomedical research.

Meet the group ↗   Research opportunities ↗

Recent publications

2026

REEL: Brain-Guided Video Generation and Selection by Reranking Encoder Embeddings with Linear Decoding from fMRI data

arXiv

Visual brain decoding from functional magnetic resonance imaging (fMRI) data has made remarkable progress in recent years, driven by increasingly powerful artificial intelligence (AI) models and the growing availability of large-scale public datasets. However, despite the impressive advances achieved in static image decoding, the reconstruction of videos fro…Read the full abstractCollapse abstract
Visual brain decoding from functional magnetic resonance imaging (fMRI) data has made remarkable progress in recent years, driven by increasingly powerful artificial intelligence (AI) models and the growing availability of large-scale public datasets. However, despite the impressive advances achieved in static image decoding, the reconstruction of videos from fMRI signals remains challenging due to the temporal mismatch between rapidly changing video stimuli and the slow dynamics of the blood-oxygen-level-dependent (BOLD) response. We evaluate a Bayesian-inspired proposal-and-rerank strategy on CC2017 and the BOLD Moments Dataset (BMD), using a linear semantic decoder, a pretrained video generator, and three neural encoding streams with ridge readouts and learned pooling in two streams. We refer to this framework as REEL (Reranking Encoder Embeddings with Linear decoding). Compared with a single generated candidate, reranking increases mean Video 50-way accuracy from 0.2359 to 0.2434 across three CC2017 participants and from 0.3653 to 0.3920 across ten BMD participants. These findings suggest an alternative and conceptually simpler strategy for improving video brain decoding, highlighting the potential of our approach as a complementary direction to increasingly complex pipelines.
2026

EXPRESS: Blood-free full quantification of dynamic TSPO PET imaging with [11C]PBR28

Journal of cerebral blood flow and metabolism : official journal of the International Society of Cerebral Blood Flow and Metabolism

BACKGROUND: Quantitative dynamic PET imaging requires metabolite-corrected arterial input functions, traditionally obtained through invasive arterial blood sampling, limiting broader adoption of quantitative kinetic modelling. Achieving generalisability across datasets, scanners, and imaging centres remains a major challenge for data-driven alternatives. ME…Read the full abstractCollapse abstract
BACKGROUND: Quantitative dynamic PET imaging requires metabolite-corrected arterial input functions, traditionally obtained through invasive arterial blood sampling, limiting broader adoption of quantitative kinetic modelling. Achieving generalisability across datasets, scanners, and imaging centres remains a major challenge for data-driven alternatives.

METHODS: We developed ASTRA, a deep-learning-based framework for arterial input function estimation from dynamic brain TSPO PET. ASTRA was developed and validated using arterial measurements in one dataset (n=90), then transfer-learned and evaluated against arterial measurements at a second centre (n=52). Building upon this, ASTRA was applied to an independent clinical cohort (n=32) from the second site but without arterial reference data, where quantitative analyses had therefore not been performed, to assess whether it could reproduce previously reported semi-quantitative cross-sectional group differences. Robustness to altered tracer kinetics was evaluated in an out-of-distribution pharmacological blockade cohort (n=5).

RESULTS: ASTRA-derived quantitative estimates agreed with arterial measurements (VT: CCC=0.658-0.805), with higher uncertainty associated with greater error. In the clinical cohort, ASTRA reproduced SUVR-based group differences using DVR (p<0.001), except in the pituitary gland (p=0.12). Finally, ASTRA yielded receptor occupancy estimates (0.92±0.17) closely matching arterial values (0.94±0.12) in the blocking cohort.

CONCLUSION: These findings demonstrate the feasibility of multi-site, non-invasive quantitative TSPO PET.
2026

Direct Conditioning of Audio Diffusion Transformers on fMRI Reveals Cortical Contributions to Sound Reconstruction

arXiv

Reconstructing audio from brain activity remains challenging. Existing brain-to-audio pipelines typically map neural responses to predefined stimulus representations before synthesis. While recent work in the visual domain has explored direct conditioning on brain activity, this approach has not yet been extended to audio diffusion models. We introduce MAAS…Read the full abstractCollapse abstract
Reconstructing audio from brain activity remains challenging. Existing brain-to-audio pipelines typically map neural responses to predefined stimulus representations before synthesis. While recent work in the visual domain has explored direct conditioning on brain activity, this approach has not yet been extended to audio diffusion models. We introduce MAAS (Multi-Area Auditory Synthesis), an end-to-end framework that conditions a pretrained audio diffusion transformer on ROI-wise fMRI responses without first supervising a decoder to predict an external stimulus representation. MAAS uses region-specific projections and learned brain tokens to condition Stable Audio through a dedicated cross-attention pathway. On a 7T fMRI dataset, we demonstrate consistent improvements over baselines, including two-stage fMRI-to-latent conditioning, across both acoustic and semantic evaluation metrics. In addition, we introduce a fidelity metric grounded in a computational model of auditory neural processing, which allows us to quantify the contribution of individual auditory cortical regions to reconstructed audio representations.

Latest 50 publications ↗