Color flow-directed doppler studies of ovarian masses: Computer analysis

OBJECTIVE: To determine if analyzing the entire color Doppler image (CDI) pulse wave of an ovarian mass can improve the ability to predict its histopathology. STUDY DESIGN: The CDI of 42 histopathologically diagnosed adnexal masses were retrospectively analyzed. Using an image analysis software program, the following parameters were calculated: area under the curve, compactness, Feret diameter, perimeter, shape factor and width of the waveform. Using an automated curve-fitting software program, the up and down slopes were processed separately for the optimal equation and coefficient for each slope. Two computerized neural networks were created, both consisting of an input layer, one hidden layer and an output layer of three neurons: benign, borderline and malignant. The first network contained two input neurons: pulsatility index (PI) and resistance index (RI). The second network contained 10 input neurons consistent with the shape and slope parameters calculated. The coefficient of determination (R2) was determined for each network. RESULTS: The neural network utilizing RI and PI failed to train (1,397 runs, 67,056 facts, R2 = 0.59, 0.12 and 0.43 for benign, borderline and malignant masses, respectively). The network using the 10 calculated parameters achieved an R2 of 0.96 after 685 runs and 27 facts. CONCLUSION: Analyzing the CDI studies of ovarian masses, using the entire pulse wave, improved the ability to differentiate between their benign, borderline and malignant histopathology.