Fast,interactive algorithm for segmentation of a series of related images: Application to volumetric analysis of MR images of the heart

Magnetic resonance (MR) cine images of the beating heart have excellent spatial and temporal resolution. Extracting the boundaries of the heart from MR images for volumetric measurements is of considerable interest; however, since the number of images involved is large, tracing the boundaries by hand is tedious and prohibitively time consuming. The authors have developed an interactive method of boundary detection that uses the correlation between the cardiac boundaries on temporally or spatially adjacent images to increase the speed of the process and reproducibility of the measurement. A simulated cine MR study of a phantom (total of 155 images) and cardiac cine studies of two patients (192 images each) were analyzed by two independent observers. Analysis of the phantom data was completed in 5.6 minutes (2.16 seconds per image) by observer 1 and 6.3 minutes (2.4 seconds per image) by observer 2. The percent measurement errors for 31 phantom volumes (30-120 mL) were 0.96% and 0.83% for observers 1 and 2, respectively. The observers analyzed the patient studies in 14-23 minutes (4.4-7.2 seconds per image), with interobserver variabilities of 5.8% and 3.7% for the two patients, respectively. The authors conclude that their flexible, semiautomatic, interactive algorithm allows rapid and reproducible detection of structural boundaries.