Research On Key Technologies Of Medical Data Processing Based On Neural Network

Medical big data,as an important component of the "national big data strategy," is a crucial element in promoting the construction of a "datadriven country" and is also a vital resource related to the overall health of the population.With the continuous development of artificial intelligence technology represented by neural networks,new growth drivers and vitality have emerged in the healthcare field,leading to the rapid development of medical big data.The challenge lies in how to extract value from medical data,expand the theory and application of medical data analysis,and transform data information into strategic resources that serve national development,social progress,and the overall health of the population.This dissertation focuses on researching key technologies for medical data processing based on neural networks,to improve the disease diagnostic performance of medical data.It proposes a solution model based on neural networks.The main content and contributions of this dissertation are as follows:(1)Propose a missing data imputation model based on feature association rules.This dissertation proposes a feature association rule-based imputation model for dealing with missing data in structured medical data.The model first calculates the feature contribution to evaluate the influence of nonmissing data features on missing data features and then selects features based on their contribution.Finally,a BP neural network imputation model based on association rule learning is constructed,which can predict missing data by learning and expressing feature association rules,thus improving data completeness.Experimental results demonstrate the effectiveness of the proposed model.(2)Propose a CT-assisted diagnosis model with adaptive local feature enhancement.This dissertation presents a computer aided diagnosis model based on adaptive local image enhancement combined with object detection to address the issue of insufficient feature expression in CT image data,which leads to poor performance in computer-aided diagnosis.Firstly,the model traverses the pixels of CT images and calculates the probability density function(PDF)of pixel grayscale levels.Then,according to the number of classes of pixel gray levels calculated from the probability density distribution of pixels,the pixels are classified and each class is subjected to image enhancement based on histogram equalization.Finally,a target detection model based on YOLOv5 is established to detect and identify lesion areas.The model was experimentally validated on different datasets of stroke CT images,and the results confirmed the accuracy of the proposed model,as well as its strong generalization capability.(3)Propose a disease classification model for multimodal medical data.To address the challenges of data heterogeneity and high dimensionality in multimodal medical data,this dissertation designs a disease classification model that integrates and classifies multimodal medical data using graph data structure construction and graph learning.The study introduces the construction of graph data structure into the integration of multimodal medical data.Firstly,a method based on graph data structure construction is designed to integrate multimodal medical data.This involves feature extraction and preprocessing performed on multimodal disease data,including quantization,normalization,etc.Then,the construction of a graph data structure based on similarity,to integrate multimodal medical data.Subsequently,a graph neural network-based data learning model is established to sample and aggregate node information from the generated graph data structure,producing new node embeddings.Finally,the model completes the task of classifying nodes,achieving patient disease classification.Experimental results on two cardiac disease datasets validate the effectiveness of the model.In summary,this dissertation focuses on key technologies for medical data processing,conducting research aimed at improving the diagnostic performance of medical data.The study proposes a neural network-based method for analyzing and modeling medical data,achieving support for theoretical and technical advancements in the field of medical big data.