Title Research On Modeling Disease Process Of Alzheimer’s Disease Via Structured Multi-task Learning

Alzheimer’s disease is a neurodegenerative disease with a slow onset at first but will continue to worsen.It not only causes progressive damage to patients’ memory and other cognitive functions,but also has become a huge economic burden on the healthcare system.The two most challenging issues in this field are（1）accurate prediction of cognitive performance of brain diseases,and（2）search for imaging biomarkers related to the pathogenesis.Researchers at home and abroad have put forward many excellent forecasting models in combination with the above two challenges.However,most of these models tend to（1）use models that are not interpretable such as deep learning,（2）a singletask modeling method,without considering the potential information between the data,and（3）the assumptions of the model Too harsh or not fully considering the actual conditions that exist in the doctor’s actual diagnosis and treatment process,resulting in a good simulation effect of the model,but poor effect in real application scenarios.In order to solve the above problems,this paper introduces a multi-task learning theoretical framework.We will use structured multi-task learning to model the progression of Alzheimer’s disease.The main research work and contributions of this paper are as follows:（1）Aiming at the mindset that most current models only consider the single-task mode to predict the recognition score,a modeling method using multi-task learning is proposed.Joint task modeling is carried out by sharing information about the subjects’multiple time follow-up points for cognitive assessment.（2）The current real situation in medical diagnosis is fully considered,and more accurate and objective data sources are considered for disease progression analysis,and the experimental results prove that the program is effective and more suitable for auxiliary diagnosis and treatment.（2）Look for key biomarkers.Aiming at the limitations of the model’s lack of interpretability and being difficult to be recognized by the medical field,the use of structured-based multi-task learning algorithms not only makes the model interpretable as a whole,so that doctors can explain the reasons behind the better performance of the algorithm,but also use more the characteristics of task data sharing between tasks can make up for the problem of insufficient prediction accuracy.