Abstract:
Chronic gastritis (CG) is a highly prevalent disease of the digestive system. In the diagnosis model of traditional Chinese medicine (TCM), whether a patient has a certai...Show MoreMetadata
Abstract:
Chronic gastritis (CG) is a highly prevalent disease of the digestive system. In the diagnosis model of traditional Chinese medicine (TCM), whether a patient has a certain syndrome is determined through a combination of symptoms. However, TCM diagnosis for CG syndromes has proven quite challenging. First, due to the large number of symptoms, the correctness of diagnosis largely depends on the doctor’s experience, which might be subjective. Second, collecting all the symptoms for diagnosis in advance is time-consuming.To address the two challenges, we first design an ensemble learning model for the diagnosis of CG syndrome using LightGBM (Light Gradient Boosting Machine). The model can diagnose new CG syndrome instances effectively. We also adopt a voting mechanism with four feature selection algorithms to select the most relevant symptoms. We collected a total of 2,680 CG cases, of which 536 (20%) cases were used as a test set to evaluate the proposed model. The model with LightGBM has an average diagnosis accuracy of 91.64% for 10 syndromes, which is higher than two prevailing methods kNN and SVM. In addition, we also identified 36 most relevant symptoms for the diagnosis of Piwei Shire pattern with four feature selection algorithms. The diagnosis accuracy obtained with partial features can reach 86.75%, which is slightly higher than that of full-feature diagnosis. The results reveal that using only partial features that have the most significant impact on the diagnosis results will not reduce the diagnosis accuracy.
Date of Conference: 09-12 December 2021
Date Added to IEEE Xplore: 14 January 2022
ISBN Information: