• Language
• Autism Spectrum Disorder (ASD)
• MRI analysis

In the past 50 years, Autism Spectrum Disorder (ASD) has gone from a narrowly defined, rare disorder of childhood onset to a well publicised, advocated, and researched lifelong condition, recognised as fairly common and very heterogeneous. As recently reported, ASD has a prevalence of about one in 87 children aged 7-9 years in Italy, raising the need for improved ASD policies for children and their families in the public healthcare system. While the complex etiopathogenic mechanisms contributing to Autism remain elusive, the past decade has seen convergence across advanced neuroimaging research modalities pointing to the presence of differences in brain structure and function architecture potentially serving as biomarkers of the disorder, even translatable in clinical settings. The main purpose of the current work has been to explore the role of advanced analysis of MRI data sets in the field of Autism Spectrum Disorder, particularly focussing on recent machine learning approaches. Chapter 1 contains a systematic review of literature concerning structural MRI data in ASD subjects, discussed adopting a “machine learning perspective”. Patients with ASD were observed to have increased whole brain volume, particularly under 6 years of age, compared with typical controls. Other consistent changes in ASD included increased volume in the frontal and temporal lobes, increased cortical thickness in the frontal lobe as well as reduced cerebellum volume and reduced corpus callosum volume. Chapter 2 contains an original research study in which we analysed MRI datasets of two different clinical groups (aged 30-80 months; non-verbal IQ >70): 26 ASD subjects and 27 Childhood Apraxia of Speech (CAS) subjects and we compared them to 20 typically developing subjects (TD), after matching for age and sex. Cortical volumes, cortical thickness and other subcortical structure (extrapolated using FreeSurfer software) differentiated the three groups in regions mainly distributed within the fronto-temporale lobes, consistently with their crucial role in language skills development. Advanced machine learning analysis confirmed this result and also support a possible partial overlapped language-related background of these two conditions, in line with previous genetic studies. Finally, some methodological consideration regarding structural MRI acquisition in children with neurodevelopmental disorders are reported in Chapter 3. In particular, the sleep MRI procedure according to UC Davis MIND Institute model is discussed, a way potentially useful to enlarge clinical cohorts with minimal risk, and thus empowering meaningfull contribute to knowledge in this field.