Measurable and Unmeasurable Features of Ultrasound Lymph Node Images in Detection of Malignant Infiltration.

The aim of the study was to assess diagnostic value and utility of selected morphological features in predicting lymph node (LN) malignancy using B-mode, Doppler ultrasonography and multivariate settings in a tertiary radiological referral center. The study included 123 patients having undergone ultrasound-guided fine-needle aspiration and cytologic analysis (FNAC) of cervical, axillary and inguinal LNs. Each LN was characterized by long/L and short/T-axis, shape, margins, echogenicity, cortical thickness, vascularization, and examiner's subjective impression. Within the limitations of FNAC, altered shape and vascularization had relatively high specificity and positive predictive value (>80%), whereas subjective impression had high sensitivity and negative predictive value (100%) for malignancy. The cut-off levels for different features of LN by ROC analysis were as follows: long-axis 23 mm, short-axis 11 mm, L/T ratio 2.19, and maximal cortical thickness 5.1 mm. On multivariate analysis (adaptive regression splines, n=108), the addition of long-axis, L/T ratio, age and sex considerably improved diagnostic accuracy (88%), sensitivity (margins + vascularization) and specificity (subjective impression) of the diagnostic model. The combination of morphological and demographic features could improve diagnostic accuracy, usually with a trade-off between the sensitivity and specificity of the predictive model. The performance may depend on the level of expertise and institutional settings.