Classification of the Severity of Diabetic Autonomic Neuropathy Using SVM+

• Main Authors: Ken Ying-Kai Liao, 廖英凱
• Other Authors: Chuang-Chien Chiu
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/48490209181592798076

碩士 === 逢甲大學 === 生醫資訊暨生醫工程碩士學程 === 101 === Diabetes mellitus is a common chronic disease. Type 2 diabetes, the more common type of diabetes, usually affects older people. However, in recent years, the number of young adults being diagnosed with type 2 diabetes is also increasing. Diabetes can lead to neuropathy, by the damage of small vessels leading to the nerves. The autonomic nervous system plays a large role in keeping our body balanced, through the sympathetic and parasympathetic nervous systems. If diabetes causes damage to the autonomic nervous system, it can lead to diabetic autonomic neuropathy. The severity of diabetic autonomic neuropathy can affect the quality of life of people, and needs to be diagnosed in order to be treated properly. This study recruited 39 patients with diabetes, with the severity of autonomic neuropathy ranging from mild neuropathy to severe neuropathy. 15 age and gender matched healthy controls were also recruited. The subjects’ cerebral blood flow velocity and blood pressure were measured with a transcranial Doppler ultrasound, and a Finapres continuous blood pressure measurement system. The data were stored and processed by a computer, and were analyzed with cross-correlation function and linear regression, which have been proven to be useful tools in assessing diabetic autonomic neuropathy. The results were processed by an SVM+ classifier to build a model to classify the varying degrees of autonomic neuropathy. The SVM+ classifier uses a new learning paradigm called “learning using hidden information”. This learning paradigm allows the SVM+ classifier to be efficient by only building the model with the hidden information, and thus patients may not have to go through many difficult tests to know the severity of their diabetic autonomic neuropathy. The program was written in the LabVIEW® environment, to be compatible with previous data collection methods. Normal controls and mild and severe diabetic autonomic neuropathy subjects have been successfully classified with a 92.59% classification accuracy based on leave-one-out cross validation using an SVM+ classifier with the modified composite autonomic scores as hidden information.