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

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

In this paper, the attention deep learning algorithm is preliminarily illustrated with human hematological data, and the performance of different algorithms are compared. Compared with EHR without unified evaluation standards, this paper used human hematological parameters which had unified standards as training data, so that the algorithm has higher reliability. Our attention deep learning algorithm has been preliminarily applied to the automatic diagnosis of hyperlipidemia with hematological parameters; the corresponding diagnostic markers and the significance of different markers are given at the same time. The hematological parameters used in this study include cholesterol, triglyceride, high-density lipoprotein, low-density lipoprotein, hemoglobin and red blood cells. Despite different blood parameters are often obtained by different methods, their acquisition methods have a unified standard. These parameters can reflect different health conditions of human body. 32, 33 At the same time, compared with the auxiliary diagnostic method which only provides diagnostic results, the algorithm proposed by this paper can also predict the diagnostic markers of diseases and the corresponding importance automatically, which improve the possibility of finding more effective new diagnostic markers, and accelerate the development of medical diagnostic technology further. This paper compares the results of the model's automatic prediction with the gold standard. Although no new diagnostic markers were obtained using limited data, it proved that the model has the potential to reasonably predict the diagnostic basis. Our work is the first time to systematically study an artificial intelligence aided diagnosis system that integrates automatic diagnosis and automatic prediction diagnostic markers, which is of great significance to the development of medical level. The improvement of recognition accuracy does not mean the improvement of medical diagnosis, but also the explanation of disease mechanism. Increasing the interpretability of the model will further improve the diagnostic level of the disease. 34 In addition, a method that can automatically process a large number of samples and provide biomarkers can speed up the study of disease mechanism. In conclusion, the combination of deep learning technology and medical diagnostic technology is of great significance for disease research.