Detecting Diseases by Human-Physiological-Parameter-Based Deep Learning

• Published in: IEEE access Vol. 7; pp. 22002 - 22010
• Main Authors: Liu, Yuliang, Zhang, Quan, Zhao, Geng, Qu, Zhigang, Liu, Guohua, Liu, Zhiang, An, Yang
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Artificial intelligence; Artificial neural networks; automatic diagnosis; Blood; Confidence intervals; Convolution; Data models; Deep learning; Diabetes; Diabetes mellitus; Diagnostic systems; Diseases; Human resources; Inspection; Mathematical models; Medical diagnostic imaging; Neural networks; Parameters; physiological parameters of human; Physiology
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2019.2893877

The application of artificial intelligence in auxiliary diagnosis diseases has become a current research hotspot. The traditional method of diagnosing diabetes circulatory complication, diabetic peripheral neuropathy hyperlipidemia, diabetes mellitus peripheral angiopathy, and the comprehensive diseases is to distinguish an inspection report by a professional doctor. Its implementation of the clinical decision support algorithm for medical text data faces a challenge with the confidence level and accuracy. We proposed an expanding learning system to detect diseases above in our medical text data, which cover many kinds of physiological parameters of human, such as hematologic parameters, urine parameters, and biochemical detection. First, the raw data were expanded and corrected. Second, the processed data were fed into a 1D-convolution neural network with dropout and pooling. Our algorithm achieves 80.43%, 80.85%, 91.49%, 82.61%, and 95.60% with testing datasets (46 subjects). The effect of data quantification on model performance also had been researched, and the different data quantification methods would affect model performance on distinguishing different diseases. The proposed auxiliary diagnostic systems that have a highly accurate and robust performance can be used for preliminary diagnosis and referral; therefore, it is not only saving many human resources but also resulting in improved clinical diagnostic efficiency.