The aim of the study is to compare the effects of 2 different dosages and modalities of motor-cognitive rehabilitation in Parkinson's disease with postural instability and gait disorders (PD-PIGD) on clinical features, neuroimaging and blood-based biomarkers at short-term (2 months) and long-term (7 months) follow-up. Fifty subjects with PD-PIGD will be randomized in 2 training groups: DUAL-TASK+AOT-MI and the DUAL-TASK groups. The DUAL-TASK+AOT-MI group will perform a dual-task gait/balance training consisting of action observation training (AOT) and motor imagery (MI) combined with practicing the observed-imagined exercises; DUAL-TASK group will perform the same exercises combined with watching landscape videos. The training will last 6 weeks, 3 times/week, 1 hour per session. Before and after training (W6), all the patients will undergo neurological, gait/balance, cognitive/behavioral, magnetic resonance imaging (MRI) and serum biomarkers evaluations. Neurological, gait/balance, cognitive/behavioral assessments and serum biomarkers will be also repeated at the 14-week follow-up (W14) to assess maintenance of results. Patients of both DUAL-TASK+AOT-MI and DUAL-TASK groups will be further randomized to repeat the training (6 weeks, 3 times/week, 1 hour each session) starting at W14 (DUAL-TASK+AOT-MI\_DOUBLE and DUAL-TASK\_DOUBLE groups). After six weeks (W20) all the subjects repeating the training will be evaluated (neurological, gait/balance, cognitive/behavioral assessments). At 28-week follow-up (W28), the whole sample of patients will be assessed with neurological, gait/balance, cognitive/behavioral, MRI and serum biomarkers evaluations. All MRI scans will be acquired at least 12 hours after last dopaminergic therapy administration to mitigate the pharmacological effects on neural activity. Twenty age- and sex-matched healthy controls will be recruited to perform gait/balance and cognitive/behavioral assessments, blood sample and brain MRI acquisition at baseline. The secondary aims of the study are to define the neuroimaging and blood-based biomarkers of PD-PIGD patients presenting different clinical features (e.g. presence of mild cognitive impairment, freezing of gait, falls and mood disturbances) and to evaluate the role of blood-based and neuroimaging biomarkers, together with clinical characteristics, in predicting the response to different dosages of rehabilitation in PD-PIGD throughout the development of a machine-learning algorithm.