FMRI Data Analysis Based On Constrained NMF And Application Research On Plasticity In Seafarer

Transportation is an important foundation for promoting the construction of China’s Belt and Road Initiative,in which the shipping industry plays a key role in the transportation industry and contributes positively to the development of the national economy and foreign trade.With the development of artificial intelligence and other emerging technologies,the shipping industry is gradually moving towards the era of intelligence.Facing the new changes brought by intelligent shipping,the cultivation of high-quality shipping talents in line with the market demand is a common problem faced by the development of the global shipping market.The challenge of modern navigation’s high demand for seafarers’ comprehensive quality requires the integration of new selection and training methods and the provision of neurophysiological support for seafarers’ training,to adapt to the demands of the shipping industry for seafarers’ talent training in the context of the era of rapid development of artificial intelligence,and thus to further enhance the safety level of intelligent shipping.The brain is the most complex system in nature,integrating a series of high-level cognitive functions and endowing human beings with special intelligence,the core of which is learning ability and plasticity.Among various brain research methods,functional magnetic resonance imaging(fMRI)has become an important means to study brain function since its emergence due to its superior characteristics.Using this technique,brain activity can be monitored to determine the cognitive,emotional,and behavioral characteristics of an individual,and thus more accurately assess the occupational plasticity of seafarers and guide the selection and training process.However,there are limitations in the current conventional methods of brain functional connectivity testing using fMRI technology.Therefore,it is essential to develop and refine brain functional connectivity assays to facilitate the exploration of neural mechanisms in the brain and the study of occupational plasticity in seafarers.The non-negative Matrix Factorization(NMF)model has been widely used in the field of fMRI data analysis due to its simplicity,interpretability,and feature focusing,and is a mainstream method for brain functional connectivity detection at present.As a datadriven blind source signal separation technique,NMF has its limitations in the fMRI data analysis process,such as the number of components being difficult to accurately estimate,the decomposition components being disordered and the time complexity being high.To address the above problems,some studies have proposed to add some available a priori information into the estimation process of traditional NMF and develop it into the constraint non-negative matrix factorization(CNMF)method,which can improve the performance of NMF in the analysis of blind source signals,and promote the study of the functional connectivity of the fMRI brain.brain functional connectivity.However,there are still many problems in the existing CNMF methods,such as the difficulty of obtaining a priori information,the difficulty of selecting a priori constraints,and the low accuracy of a priori information,which need to be further investigated and solved.To this end,the following research work is mainly carried out in this paper:Aiming at the current difficulties in obtaining intrinsic a priori information from fMRI data and the problems of large data sample size and high time complexity in group fMRI data analysis,a group spatial constraint non-negative matrix factorization(SCNMF)method was proposed to mine valuable spatial a priori information in the group subjects to guide the analysis of the group-subject level fMRI data.The components of interest were selected by performing NMF analysis at the single-subject level.Subsequently,valuable spatial a priori information was mined at the group-subject level and incorporated into the estimation process of the CNMF objective function to improve the efficiency of the algorithm.The results demonstrated that the method can effectively improve the spatial reproducibility of the algorithm and provide new insight into parsing fMRI data.Immediately after that,for the sake of reducing the accuracy requirement of the CNMF method on the a priori information and improving its fault tolerance,a spatiotemporal constraint non-negative matrix factorization(STCNMF)algorithm based on the real reference signal was further proposed.By introducing temporal a priori information and spatial a priori information,to enhance the algorithm’s fault tolerance and spatial recovery ability.The method was also applied to extract large-scale brain networks to identify brain function information of neurocognitive networks associated with multiple real datasets.The results indicated that the algorithm had better fMRI source signal recovery performance,which helped to discover the changes in brain functional information in multiple data sets,and provided strong support for understanding the potential biomarkers in real datasets.Further,to avoid the limitations caused by model setting restrictions,this paper systematically investigated the constraints existing in current CNMF methods and proposed a multi-constraint non-negative matrix Factorization(MCNMF)algorithm.Based on the actual situation of the model inputs,the method performed sparsity constraints and orthogonality constraints on the decomposition factors,and at the same time combined the spatio-temporal a priori information to minimize the redundancy between different bases and reduce the complexity of the algorithm.Based on multiple real data sets,the performance advantages and disadvantages of the MCNMF algorithm are verified,compared,and analyzed,and the results showed that the method further enhanced the performance of fMRI data analysis,effectively identified the alterations of brain functional information in multiple real data sets at different scales,and provided neuroimaging evidence for the differential diagnosis of the specific features of each group of subjects.Finally,the MCNMF fMRI data analysis method proposed in this paper was applied to conduct an in-depth study of seafarer plasticity.To deepen the understanding of the influence of occupational factors on seafarers’ neuroplasticity,and to solve the problem that the current research only focuses on a single group of subjects or a single scale,a twotopological scale hierarchical analysis framework based on the MCNMF algorithm was developed,to explore the public and specific connectivity patterns of pre-seafarer,backseafarer,and healthy control groups from different scales.The results of the study revealed that: public connectivity patterns existed at different scales in pre-seafarer and backseafarer;pre-seafarer and back-seafarer had their idiosyncratic connectivity patterns;and seafarers’ subjects’ connectivity properties had a tendency to change at different hierarchical scales.The results of these studies increased the understanding of the relationship between seafarers’ functional brain plasticity and occupation of seafarers,provided neurophysiological support for subsequent seafarer selection and training,and promoted the quality of nautical education and training as well as the safety level of shipping traffic.