Research On Feature Learning Model Of Medical Data

The big data applications in field of healthcare is a part of national strategic layout of big data,which is also related to the development of national healthcare.The big health data has the properties of multi-modal,sustained and fast growth,and complexity,bringing great challenges to the analysis and application of the data.There are many crucial issues to be addressed in current large health data analysis and application,like how to timely and accurately collect and obtain health data,how to efficiently use the high-speed networks to obtain the reliable transmission of information(digital,image,audio etc.),how to dig out useful information from the large health care data with artificial intelligence technologies,and how to develop the intelligent applications for majority of medical workers and ordinary people.With respect to the problem of data mining and application development about big health care data,this paper studies the feature representation learning methods of multi?modal health data and proposes several feature learning models applying to the disease risk assessment.Regarding to the issues of providing sound data services for intelligent medical applications,cloud-end fusion based semi-physical simulation method is proposed.The main contributions of this paper are as follows:(1)In the aspect of medical text feature learning,this paper proposes a medical text feature learning model based on convolutional neural network.Deep learning is adopted in the text analysis of disease risk assessment application,in which text feature representation is realized with deep learning methods and the feature learning and extraction of different diseases are implemented with the same method to realize the universality of the model.In order to reduce the dependence of model on text data,the structured feature designed by medical professionals is added,and an assessment model fusing medical text data feature and structured feature is proposed.The experimental results show that the model is effective,universal and stable.(2)In the aspect of multi-dimensional data feature learning,a tensor-based convolutional auto-encode neural network(TCANN)model is proposed.Compared with usual image data,the medical image data is multi-dimensional,whose pixel space contains more information.Tensor computation is applied to the TCANN model,extending the model application from vector space to tensor space.And the tensor distance is used as the error function to capture the distribution characteristics of data in tensor space.The proposed model is used to assess the risk of lung nodules based on the lung CT(Computed Tomography)images and the retrieval of similar nodules,which has be proven to be effective.(3)In the aspect of learning features of time series data,a multi channel convolutional auto-encode neural network(MCAE)is proposed.The model analyzes the relationship between fatigue and emotional abnormality,puts forward the concept of emotional fatigue.In the proposed model,the feature of ECG is learned by MCAE,the feature of facial images is learned by convolutional auto-encode neural network,then the emotional fatigue is detected by fusing these two learned features and acquired physiological signals.Besides,an emotional fatigue detection platform based on multimodal data fusion is established,which realizes the function of data acquisition,emotional fatigue detection and emotion feedback.The platform is tested and verified that the model proposed can be used to learn features of time series data(4)In the aspect of providing services for intelligent medical applications,a cloud-end fusion based semi-physical simulation method is proposed.In intelligent medical applications,the data has the properties of diversity and complexity.In order to research on how to build a flexible network architecture and provide safe and effective data transmission,the paper built a cloud-end fusion based semi-physical simulation model towards intelligent healthcare.The model provides technical support and reference for constructing a real cloud-end fusion based system.