Malignancies of the nasal cavity and paranasal sinuses account for 1% of all malignancies and 3% of malignancies of the upper aerodigestive tract. In the sinonasal tract, nearly half of all malignancies arise in the nasal cavity, whereas most of the remaining malignancies arise in the maxillary or ethmoid sinus. Squamous cell carcinoma is the most common histological subtype of malignant tumors occurring in this area, followed by other epithelial carcinomas, lymphomas, and malignant soft tissue tumors. Although many of these tumors present with nonspecific symptoms, each tumor exhibits characteristic imaging features. Although complex anatomy and various normal variants of the sinonasal tract cause difficulty in identifying the origin and extension of large sinonasal tumors, the invasion of vital structures such as the brain, optic nerves, and internal carotid artery affects patients’ prognosis. Thus, diagnostic imaging plays a key role in predicting the histological subtype and in evaluating a tumor extension into adjacent structures. This article describes the computed tomography and magnetic resonance imaging findings for malignant sinonasal tumors.

SUMMARY: We present 4 cases of carcinoma ex pleomorphic adenoma of the parotid gland. In 3 of the 4 cases, diffusion-weighted and apparent diffusion coefficient (ADC) mapping images clearly revealed carcinoma as a hypercellular area with low ADC values and pleomorphic adenoma as a hypocellular area with high ADC values. Diffusion-weighted images demonstrated well complex tissue components in carcinoma ex pleomorphic adenoma, which may be useful for the diagnosis of this disease.

OBJECTIVE. The purpose of our study was to preliminarily evaluate the usefulness of a computer algorithm analysis using the finite difference method and an artificial neural network to diagnose hepatic fibrosis with MR images.MATERIALS AND METHODS. Liver parenchymal textures on the MR images of 52 patients who underwent partial hepatectomy were processed by the computer algorithm and reviewed by two radiologists. The texture features using the finite difference method were processed by an artificial neural network program containing a three-layer learning algorithm of the back propagation, composed of a seven-unit input layer, a six-unit hidden layer, and a one-unit output layer. The radiologists assigned confidence levels for the presence of hepatic fibrosis. Degrees of hepatic fibrosis were determined semiquantitatively by a pathologist. Algorithm outputs and radiologists' interpretations were correlated with degrees of fibrosis using Spearman's rank correlation analysis, and diagnostic performances wer...