Decision support system with the structural equation modeling for prediction of malignancy on thyroid nodes that are diagnosed as atypia of undetermined significance or follicular lesion of undetermined significance

The patient group in which decision-making in thyroid cancers is the most complex is the patients diagnosed with AUS/FLUS in the 3rd category of the Bethesda System. Fine-needle aspiration biopsy (FNAB) is frequently recommended for such patients, whose malignancy risk ranges from 10-30%, and sometimes results in surgical intervention upon the patient's request. In the study, patients with nodules who applied to a clinic with the suspicion of thyroid cancer and defined in AUS/FLUS type cytopathology as a result of FNAB; A clinical decision support system based on the most successful model has been proposed to assist the physician in determining the benign/malignant status of preoperative nodules using artificial intelligence and machine learning methods using clinical information, ultrasonography and cytopathological findings. In the aforementioned data set, the structural equation modeling method was used in order to determine the relationships between the variables and to make statistically significant estimations. On the meaningful variables determined by this method, support vector machines (SVM), which is one of the machine learning / artificial intelligence approaches, Naive Bayes, a probabilistic classifier based on Bayes' theorem with independent assumptions, and decision tree methods are used to create a decision support system. As a result of the study, the sensitivity of the most successful model developed with the variables obtained using the SEM analysis was obtained as 70%, the selectivity as 93% and the accuracy as 82%. Apart from that, many models were created for different scenarios and the relevant results were compared. In conclusion; It was determined that the most successful proposed SVM model had the highest accuracy in predicting benign nodules. Thus, the decision-making process of physicians can be supported by offering a second clinical option before surgery.