Resting-state fMRI (rs-fMRI) is a widely used technique to investigate the residual brain functions of patients with Disorders of Consciousness (DoC). Nonetheless, it is unclear how the networks that are more associated with primary functions, such as the sensory–motor, medial/lateral visual and auditory networks, contribute to clinical assessment. In this study, we examined the rs-fMRI lower-order networks alongside their structural MRI data to clarify the corresponding association with clinical assessment. We studied 109 chronic patients with DoC and emerged from DoC with structural MRI and rs-fMRI: 65 in vegetative state/unresponsive wakefulness state (VS/UWS), 34 in minimally conscious state (MCS) and 10 with severe disability. rs-fMRI data were analyzed with independent component analyses and seed-based analyses, in relation to structural MRI and clinical data. The results showed that VS/UWS had fewer networks than MCS patients and the rs-fMRI activity in each network was decreased. Visual networks were correlated to the clinical status, and in cases where no clinical response occurred, rs-fMRI indicated distinctive networks conveying information in a similar way to other techniques. The information provided by single networks was limited, whereas the four networks together yielded better classification results, particularly when the model included rs-fMRI and structural MRI data (AUC = 0.80). Both quantitative and qualitative rs-fMRI analyses yielded converging results; vascular etiology might confound the results, and disease duration generally reduced the number of networks observed. The lower-order rs-fMRI networks could be used clinically to support and corroborate visual function assessments in DoC.

Mutations of the voltage-gated sodium channel SCN1A gene (MIM#182389) are among the most clinically relevant epilepsy-related genetic mutations and present variable phenotypes, from the milder genetic epilepsy with febrile seizures plus to Dravet syndrome, a severe developmental and epileptic encephalopathy. Qualitative neuroimaging studies have identified malformations of cortical development in some patients and mild atrophic changes, partially confirmed by quantitative studies. Precise correlations between MRI findings and clinical variables have not been addressed. We used morphometric methods and network-based models to detect abnormal brain structural patterns in 34 patients with SCN1A-related epilepsy, including 22 with Dravet syndrome. By measuring the morphometric characteristics of the cortical mantle and volume of subcortical structures, we found bilateral atrophic changes in the hippocampus, amygdala, and the temporo-limbic cortex (P-value < 0.05). By correlating atrophic patterns with brain connectivity profiles, we found the region of the hippocampal formation as the epicenter of the structural changes. We also observed that Dravet syndrome was associated with more severe atrophy patterns with respect to the genetic epilepsy with febrile seizures plus phenotype (r = −0.0613, P-value = 0.03), thus suggesting that both the underlying mutation and seizure severity contribute to determine atrophic changes.

Focal cortical dysplasia (FCD) type II is a major cause of drug-resistant epilepsy, often curable only by surgery. Despite its clinical importance, the diagnosis of FCD is very difficult in MRI because of subtle abnormalities, leading to misdiagnosis. This study investigates the use of 3D convolutional neural networks (3D-CNNs) for FCD detection, using a dataset of 170 subjects (85 FCD patients and 85 controls) composed of T1-weighted and FLAIR MRI scans. In particular, it investigates the benefits obtained from cross-modality transfer learning and explainable artificial intelligence (XAI) techniques, in particular Gradient-weighted Class Activation Mapping (Grad-CAM). ResNet architectures (ResNet-18, -34, and -50) were implemented, employing transfer learning strategies that used pre-trained weights from segmentation tasks. Results indicate that transfer learning significantly enhances classification accuracy (up to 80.3%) and interpretability, as measured by a novel Heat-Score metric, which evaluates the model's focus on clinically relevant regions. Improvements in the Heat-Score metric underscore the model's seizure zone localization capabilities, bringing AI predictions and clinical insights closer together. These results highlight the importance of transfer learning, including cross-modality, and XAI in advancing AI-based medical diagnostics, especially for difficult-to-diagnose pathologies such as FCD.