Abstract
Severe cerebrovascular disease is an acute cerebrovascular event that causes severe neurological damage in patients, and is often accompanied by severe dysfunction of multiple systems such as breathing and circulation. Patients with severe cerebrovascular disease are in critical condition, have many complications, and are prone to deterioration of neurological function. Therefore, they need closer monitoring and treatment. The treatment strategy in the acute phase directly determines the prognosis of the patient. The case of this article selected 90 patients with severe cerebrovascular disease who were hospitalized in four wards of the Department of Neurology and the Department of Critical Care Medicine in a university hospital. The included cases were in accordance with the guidelines for the prevention and treatment of cerebrovascular diseases. Patients with cerebral infarction are given routine treatments such as improving cerebral circulation, protecting nutrient brain cells, dehydration, and anti-platelet; patients with cerebral hemorrhage are treated within the corresponding safe time window. We use Statistical Product and Service Solutions (SPSS) Statistics21 software to perform statistical analysis on the results. Based on the study of the feature extraction process of convolutional neural network, according to the hierarchical principle of convolutional neural network, a backbone neural network MF (Multi-Features)-Dense Net that can realize the fusion, and extraction of multi-scale features is designed. The network combines the characteristics of densely connected network and feature pyramid network structure, and combines strong feature extraction ability, high robustness and relatively small parameter amount. An end-to-end monitoring algorithm for severe cerebrovascular diseases based on MF-Dense Net is proposed. In the experiment, the algorithm showed high monitoring accuracy, and at the same time reached the speed of real-time monitoring on the experimental platform. An improved spatial pyramid pooling structure is designed to strengthen the network's ability to merge and extract local features at the same level and at multiple scales, which can further improve the accuracy of algorithm monitoring by paying a small amount of additional computational cost. At the same time, a method is designed to strengthen the use of low-level features by improving the network structure, which improves the algorithm's monitoring performance on small-scale severe cerebrovascular diseases. For patients with severe cerebrovascular disease in general, APACHEII1, APACHEII2, APACHEII3 and the trend of APACHEII score change are divided into high-risk group and low-risk group. The overall severe cerebrovascular disease, severe cerebral hemorrhage and severe cerebral infarction are analyzed, respectively. The differences are statistically significant.