Identification of Brain Structures and Functional Cortico - Muscular Networks: Machine Learning Object Recognition and Network Physiology Approach

Abstract

The brain is the most complex part in the human body. This organ is responsible for our intelligence, interpreting sensation, initiating body movement, and controlling all of our behaviors. Over hundreds of years, scientists have learned much about the brain, from a microscopic and macroscopic point of view. We now know the general rules under which information is transferred from neuron to neuron and we can differentiate between various brain structures and brain areas, each of them responsible of a particular function in the human organism. However, due to the vast complexity of the brain, much remains to be discovered. Researchers continue to explore the mechanics regulating a healthy brain that functions quickly and automatically, but we are still at the point where much work remains to identify the key differences between a physiological and a pathological situation in anatomic brain structures and functionality of the brain. The lack of information in this sense affects the diagnostic process of many neurodegenerative disorders, that can be discovered only from the symptoms shown by the subject and that, therefore, can be treated to reduce the pain and to give better conditions of life. The present research aims to better understand anatomic brain structures and functional interactions networks in the brain in order to early diagnose the most common neurodegenerative diseases. In the framework of the investigation of the anatomic brain structures the Neuroimaging is the most powerful tool used in basic research and clinical field. The Magnetic Resonance Imaging (MRI) is one of the most recent techniques of brain imaging and largely used for its low degree of invasion in the human body. It can provide valuable information in the detection of morphological markers that can highlight on the healthy status of the subject. A fundamental step in the pre-processing and analysis of magnetic resonance images is the individuation of the Mid-Sagittal Plane (MSP), where the mid brain is located, in order to set a coordinate reference system for the MRI scan images, and to precisely measure small changes in the surfaces, volumes and distances between different brain areas, which are used as biomarkers in the diagnostic process of certain diseases, such as Parkinson, Alzheimer, Progressive Supra-Nuclear Palsy. In this regard, part of the present research involves the improvement of brain MRIs analysis, with the use of machine learning techniques applied for the automatic identification of the MSP. In particular, the proposed method, Image Pixel Intensity (IPI) algorithm, is implemented in MatLab and is based on the k-mean, which allow to automatically segment the 2D MRIs in different brain tissues, and automatically identifies the slice where the brain tissues are most distinct from each other exploiting the intensity of the resonance signal expressed in the MRI by the color of the grayscale pixels. The results of this algorithm have been compared with the evaluation of four medical experts who manually identified the Mid-Sagittal, providing an average percentage error of 1.84%, and demonstrating that the proposed algorithm is promising and could be directly incorporated into larger diagnostic support systems. Following the main aim of the present research, the early diagnosis of neurodegenerative diseases, another machine learning technique, elastic net, has been implemented in Matlab in order to automatically predict the brain age, exploiting relationships involving the amount of gray matter present in the brain of the subjects analyzed, through a structural MRI study. The outcome of this work is the identification of profound correlations between the expected brain age and the general cognitive state: semantic verbal fluidity, processing speed, visual attention and cognitive flexibility, and visual attention and cognitive flexibility. Among the neurodegenerative diseases Parkinson lately acquired particular interest, due to its growing diffusion even within forty years old patients. This led to the study of functional interactions networks between the brain and the locomotor system during different sleep stages. Electroencephalography (EEG) and electromyography (EMG) data of healthy subjects and Parkinson's patients have been analyzed highlighting the correlations between different frequency bands present in the electrical signals emitted in the different brain areas and in the muscles of the chin and leg. Synchronous bursts in electrical activity signals in the brain and muscles have been analyzed, using the innovative method of Time Delay Stability (TDS), based on the cross-correlation function in consecutive time windows between two different signals. Links between the different frequency bands of different brain areas and the muscles with a long stable delay of the peak in the cross-correlation function are considered more stable, then stronger. The same analysis has been conducted on healthy and Parkinson's subjects, showing substantial differences in the networks of cortico-muscular interactions involving different frequencies between a physiological situation and a pathological one. Each sleep stage is uniquely identified by a particular pattern in the brain-muscle interactions. For Parkinson’s subjects these functional patterns change during each sleep stage; moreover, in general the strength of the links decreases during wake and light sleep but increases or remains the same during REM and deep sleep, especially for the brain-leg interactions, showing that during the waking phase the brain is not able to adequately control the muscles of the lower limbs. Analyzing in details the behavior of muscles the electric activity of different muscle fibers has been studied, considering subjects of different age groups (children, young adults and elderly subjects) in situations of stress or rest. In particular, EMG signals from the muscles of the leg and the back have been taken into account. The analysis shows that rest and stress have very different patterns, due to the different types of muscle fibers involved and how they behave during muscle relaxation and contraction; these relationships also change with age, identifying patterns that uniquely identify the age of the subjects analyzed and also vary during the same exercise by marking the precise point at which the subject reaches fatigue first and exhaustion afterwards.