Research On Interpretable Prediction Model Of Traumatic Hemorrhagic Shock Based On Improved ANN

Traumatic hemorrhagic shock is one of the most life-threatening complications caused by severe trauma,with rapid onset and high mortality.If patients with traumatic hemorrhagic shock can be identified in time and accurately,and timely and effective interventions can be taken,the mortality and disability rate of patients can be greatly reduced.In the medical field,every decision is related to the life and health of the patient.To provide medical staff with reliable auxiliary decision support,it is of great significance to improve the performance of the model and enhance the interpretability of the model.Therefore,the research used the idea of multiobjective optimization to provide a new solution for the prediction of traumatic hemorrhagic shock.First of all,609 patients were extracted from the emergency database of the Chinese People’s Liberation Army General Hospital based on the inclusion and exclusion criteria under the guidance of professional clinicians,including 107 in the experimental group and 502 in the control group.Data cleaning,missing value imputation,standardization and normalization and other preprocessing processes were carried out.Three algorithms,Lasso,Random Forest and XGBoost,are utilized as the basic learning machine,integrated with the idea of ensemble learning.To improve the interpretability and stability of feature selection algorithm,the stability of each algorithms is weighted.Nine key indicators associated with traumatic hemorrhagic shock were identified: heart rate,systolic blood pressure,international standardized ratio,diastolic blood pressure,thrombin time,total carbon dioxide,hemoglobin,white blood cells and lactic acid.Secondly,the cellular genetic algorithm was employed to optimized the structure of artificial neural network.With accuracy and interpretability(network complexity)as the objective function respectively,two prediction models of traumatic hemorrhagic shock were constructed,respectively.The results suggested that,to a certain extent,the two models could meet the different decision-making preferences of medical workers,but the accuracy and interpretability of the models are not taken into account simultaneously.Furthermore,a multiobjective optimized neural network prediction model for traumatic hemorrhagic shock was constructed.Introduce multi-objective optimization,and take model accuracy and complexity as optimization objectives at the same time.The cellular genetic algorithm is used to optimize the neural network structure,and the following optimization work has been done: 1)The neural network structure is optimized by the cellular genetic algorithm;2)The neuron structure is described by qubit coding;3)The three-dimensional cellular network is established Space;4)"Breakpoint reconnection" to individual neighbors in the population;5)Cellular genetic algorithm optimization by means of improving neural network weight evolution methods.Finally,Three prediction models with different accuracy and interpretability are produced(the accuracy of the model is up to 88.6%,and the complexity is as low as 10),and the corresponding prediction rules are extracted from them,which meets the different decision-making preferences of medical workers.In summary,the prediction method of traumatic hemorrhagic shock from the perspective of deep learning interpretability was studied,which enriches the research of the severe trauma field.The set of key indicators can not only accelerate the speed of prediction,but also further explore and understand the pathophysiological mechanism of traumatic hemorrhagic shock,thus promote the development of research on the treatment of this complication.Moreover,the multiple prediction models and rules of traumatic hemorrhagic shock with different accuracy and interpretability established to assist medical workers with different preferences in decisionmaking according to actual needs.