Dipartimento di Ingegneria informatica, automatica e gestionale

Functional ultrasound imaging (fUSI) has emerged as a promising non-invasive neuroimaging modality that leverages neurovascular coupling to capture hemodynamic changes associated with neuronal activity. This study investigates the feasibility of fUSI for brain state classification in both healthy and pathological conditions using computational simulations based on the Wilson-Cowan neural mass model. Simulated electrophysiological signals were transformed into fUSI-like traces via convolution with an experimentally derived hemodynamic response function (HRF). Pathological conditions representing Alzheimer's disease (AD) and epilepsy were introduced by modifying excitatory-inhibitory balance parameters. A one-dimensional convolutional neural network (1D-CNN) was trained to classify healthy and pathological states based on either raw electrophysiological data or simulated fUSI signals. Results indicate that while classification performance was superior for electrophysiological data due to its finer temporal resolution, fUSI signals retained sufficient discriminative information for reliable classification, particularly in binary tasks distinguishing control from pathological conditions. Performance was found to be dependent on the progression of pathology, with more severe alterations leading to improved classification accuracy. However, the increased sensitivity of fUSI signals to noise led to some decline in classification performance under high noise conditions. In multiclass tasks distinguishing connectivity states within the same pathology, fUSI exhibited reduced accuracy, indicating challenges in capturing finer network-level distinctions. Clinical relevance—These findings support the potential of fUSI as a cost-effective alternative to traditional neuroimaging for detecting pathological brain states. While some challenges remain in resolving fine-grained connectivity states, fUSI could provide valuable insights into neurological disorders such as Alzheimer's disease and epilepsy, particularly when combined with advanced noise reduction and signal processing techniques.