A pilot study for identifying at risk thyroid lesions by means of a decision tree run on clinicocytological variables.

Fine-needle aspiration biopsy (FNAB) is safe, inexpensive, minimally invasive, and highly accurate in the diagnosis of nodular diseases of the thyroid. However, FNAB does not provide a reliable benign versus malignant diagnosis for 100% of the cases analysed. It is possible to increase the accuracy of the cytological diagnosis by means of information contributed by different clinical variables. In the present study we evaluate the diagnostic value of 10 variables in addition to FNAB on a series of 218 specimens for which we obtained histological diagnoses including 37 cancers (17%). The diagnostic information contributed by these variables was analyzed by means of the Decision Tree technique, an artificial intelligence-related method which forms part of the Supervised Learning algorithms. The results show that Decision Trees enable some subpopulations of patients with uncertain FNAB results to be characterized.