Intelligent Detection And Medication Rule Analysis Of Nasopharyngeal Carcinoma Based On Machine Learning

Nasopharyngeal carcinoma(NPC)is the most common malignant tumor in the head and neck.Due to its hidden location,early symptoms are not obvious and it is difficult to distinguish from Chronic Rhinosinusitis(CRS),and other benign diseases,resulting in the early detection being not easy.The treatment of NPC mainly relies on radiotherapy,chemotherapy and adjuvant therapy of traditional Chinese medicine.Due to the characteristics of traditional Chinese medicine,the medication regimen is complex and changeable,and there is no general medication rule to guide practice.To solve the above problems,the thesis proposed a machine learning algorithm to achieve early screening of NPC only through conventional biochemical indicators,and to complete the analysis of medication rules during its rehabilitation treatment.Firstly,considering the problem that NPC is not easy to detect in the early stage,conventional biochemical indicators were used to detect it through machine learning method,so as to realize the large-scale early screening of NPC.Among them,random forest(RF)performed best with accuracy,sensitivity and specificity of 93.0%,90.3% and 95.3%,respectively.Compared with the commonly methods which uses EB virus antibody detection,the accuracy was increased by 2.9% and the specificity was increased by 5.0%.The result showed that conventional biochemical indicators can be used to accurately detect NPC,and to reduce the need for expensive EB virus antibody detection.Meanwhile,the prediction accuracy,sensitivity and specificity of the method on independent sample sets were 92.5%,87.1% and 98.0%,respectively,which further confirmed the effectiveness and good generalization ability of the method.Then,considering the symptoms of NPC and CRS are similar which causes them difficult to distinguish from each other,conventional biochemical indicators are used for detection to distinguish them.Among the three classifications of NPC,CRS and control group,RF achieved the highest accuracy rate of 83.1%,but CRS detection rate was lower(69.7%).In order to improve the detection rate of CRS,the influence of different feature selection methods on the classification results were compared and analyzed,and the feature set with the best classification performance was selected.Then,we used different ensemble learning algorithms to achieve three-classification and found that stacking performs best with the CRS detection rate of 82.0%,which was 14.0% higher than that of the RF.The result demonstrated the feasibility of using conventional biochemical indicators to detect NPC and CRS.Finally,in view of the complex and changeable problem of NPC drug regimen,the drug rule analysis was carried out,and the relationship between drugs was explored.The most frequently used traditional Chinese medicine was Platycodon grandiflorum(Jie Geng).The most relevant drug pairs were Magnolia(Xin Yi)and Angelica dahurica(Bai Zhi),and the three most relevant drug groups were Fritillaria thunbergii(Zhe Bei Mu),Magnolia(Xin Yi)and Trichosanthes peel(Gua Lou Pi).Moreover,4 traditional Chinese medicines for the treatment of NPC were obtained based on Cluster analysis.The thesis focused on the detection and rehabilitation therapy of NPC,and proposed the method of combing the machine learning algorithms with conventional biochemical indicators to detect NPC,which had good detection effect,characteristic of easy implementation and low cost.The thesis analyzed the medication rule of traditional Chinese medicine for NPC,and provided the reference for traditional Chinese medicine treatment of NPC after radiotherapy and chemotherapy.In the future,it can be promoted to hospitals or internet medical software to assist doctors in making decisions.