Research On Prediction Of Hyperthyroidism Based On Deep Multi-task Learning

In recent years,an important factor affecting people’s health is the increase in chronic diseases.Chronic diseases mainly have the characteristics of slow development and staged onset.The traditional treatment-based diagnosis and treatment method cannot fully cope with this change,and thus a diagnosis and treatment method that uses early intervention for predictive prevention has emerged.Facing the increasingly complex and huge medical data,the use of deep learning technology to learn effective information and predict diseases has become one of the current research hotspots in the medical field.However,the existing research results are mostly based on medical imaging data or disease-based ICD codes the staged characteristics of the disease and the deep development laws contained in the index check sequence data during the development process have not been fully explored.In addition,the instability and other factors that may exist in the actual medical application of the prediction model are not considered.To this end,the paper focuses on the chronic disease of hyperthyroidism and studies the disease prediction model.The main contributions of the paper are as follows:(1)The deep multi-task learning disease prediction model is established.According to the clinical needs of hyperthyroidism development prediction,the paper converts multi-index problems into multi-task learning problems and proposes a disease development prediction model based on deep multi-task learning.Furthermore,the correlation between tasks in the deep multi-task model is studied,which is mainly divided into two parts.First,the paper optimizes the objective function in the deep multi-task model.By introducing class label constraints into the objective function,the basic loss of each task is assisted,and the loss between each task is balanced,thereby improving the stability of model training.Second,when judging the loss of the true value and predicted value of the indicator,the samples at the boundary of the judgment range will produce greater uncertainty in the learning convergence of the model parameters and different samples have different effects on parameter learning for different tasks.Therefore,the paper establishes a sample weight mechanism in the multi-task model to improve the model’s discriminative ability.Third,considering the value dependence between different tasks,this paper redesigns the LSTM gate structure and the cell structure as G-LSTM,and introduces the values of other tasks into the original independent and self-loop processing structure,thereby using other information of the indicator value assisted the current task learning.The validity of the model was verified by experiments with real hyperthyroidism blood test data.(2)The influence of the imprecision range of clinical laboratory test results on the prediction model is studied.First,the paper constructs different disturbance models such as Gaussian disturbance model,scale model based on the number of decimal places,original value scaling model and defines quantitative evaluation indicators.Then the paper establishes experiments and analyzes the influence mechanism of imprecision on the prediction results of the prediction model based on the experimental results.The results showed that although small imprecision has less impact on the overall prediction results,they had a higher impact on the number of individual patients.The above results are of great significance in practical medical applications.Furthermore,according to the influence mechanism,the solution to the above-mentioned problems is discussed preliminarily through multi-category fusion integration of the prediction results.Finally,the disturbance model and solution were analyzed and verified through experimental results.