Usually, physicians need radiology reports to diagnose their patients' illness. Radiology reports are mostly stored as unstructured text from which information retrieval is difficult. As a consequence, interpreting unstructured radiology reports automatically needs sophisticated text mining algorithms. In this paper, we propose a new approach for structuring textual radiology reports. In the proposed approach radiology experts provide a set of patterns each of which consists of items with specific properties. Using the patterns as class labels, a radiology report is then classified based on the items found in its sentences. In the proposed approach, the given radiology report is first classified by a Naïve Bayes classifier to support uncertainty. Then, a Boolean classifier is applied to the report which uses exact matching to achieve high precision. Finally, a Case-Based Reasoning classifier is used to learn new patterns from the report if there is any. The experimental results show that the proposed approach can dynamically classify radiology reports with high precision.