Deep Learning-Based Method of Diagnosing Hyperlipidemia and Providing Diagnostic Markers Automatically

Liu, Yuliang; Zhang, Quan; Zhao, Geng; Liu, Guohua; Liu, Zhiang

All diagnostic results were determined by an endocrinologist with 8-10 years of clinical experience. Five hundred samples were used to train model and remaining 100 samples were used to evaluate the model's performance; two parts above are independent of each other. There are 50 hyperlipidemia samples (50%) and 50 healthy samples (50%) in the testing dataset to ensure the sample balance (64 male patients (64%) and 36 female (36%) patients in the testing dataset). A completely independent testing dataset can evaluate the system's performance to identify data not in the training dataset. The raw data are multidimensional vector, it consists of hematological parameters urological parameters and doctors' diagnostic results. It is shown in Figure 5 . The raw data include blood routine parameters, biochemical test parameters, blood sugar parameters, glycosylated hemoglobin parameters and urine routine parameter. We extracted the above hematological data and diagnostic results as training vectors, and will consider adding more types of parameters in future work.