Statistical analysis of MRI parameters predicting malignancy in 141 soft tissue masses

Abstract

Since well-known grading parameters such as cellularity, mitotic rate, matrix and presence of necrosis all influence MRI signal intensity, the value of MRI in predicting malignancy is potentially high. To assess this value we studied retrospectively the findings in 141 soft tissue tumours (84 benign, 57 malignant) and evaluated a wide variety of MRI features (size, margins, signal homogeneity, shape, signal intensity, neurovascular and bone involvement, degree and pattern of enhancement and evidence of necrosis after injection of Gd-DTPA). Statistical analysis was carried out to determine accuracy of parameters individually and in combination, for predicting malignancy. Highest sensitivity was obtained for "absence of low signal intensity on T2" (100%), "mean diameter greater than 33 mm" (90%), and "inhomogeneous signal on T1" (88%). Highest specificity was obtained for "evidence of necrosis" (98%), "bone or neurovascular involvement or metastases" (94%), and "mean diameter greater than 66 mm" (87%). Association of best sensitivity and specificity was seen for "absence of low signal intensity on T2", "signal inhomogeneity on T1", and "mean diameter of the lesion greater than 33 mm" (81 and 81%).