Files and links (1)
Free to read and download, Open Access
A novel method called LOGISMOS - Layered Optimal Graph Image Segmentation of Multiple Objects and Surfaces - for simultaneous segmentation of multiple interacting surfaces belonging to multiple interacting objects is reported. The approach is based on representation of the multiple inter-relationships in a single n-dimensional graph, followed by graph optimization that yields a globally optimal solution.
Three major contributions for LOGISMOS are made and illustrated in this thesis: 1) multi-object multi-surface optimal surface detection graph design, 2) implementation of a novel and reliable cross-object surface mapping technique and 3) pattern recognition-based graph cost design.
The LOGISMOS method's utility and performance are demonstrated on a knee joint bone and cartilage segmentation task. Although trained on only a small number of nine example images, this system achieved good performance as judged by Dice Similarity Coefficients (DSC) using a leave-one-out test, with DSC values of 0.84+-0.04, 0.80+-0.04 and 0.80+-0.04 for the femoral, tibial, and patellar cartilage regions, respectively. These are excellent values of DSC considering the narrow-sheet character of the cartilage regions. Similarly, very low signed mean cartilage thickness errors were observed when compared to manually-traced independent standard in 60 randomly selected 3D MR image datasets from the Osteoarthritis Initiative database - 0.11+-0.24, 0.05+-0.23, and 0.03+-0.17 mm for the femoral, tibial, and patellar cartilage thickness, respectively. The average signed surface positioning error for the 6 detected surfaces ranged from 0.04+-0.12 mm to 0.16+-0.22 mm, while the unsigned surface positioning error ranged from 0.22+-0.07 mm to 0.53+-0.14 mm.
The reported LOGISMOS framework provides robust and accurate segmentation of the knee joint bone and cartilage surfaces of the femur, tibia, and patella. As a general segmentation tool, the developed framework can be applied to a broad range of multi-object multi-surface segmentation problems.
Following the LOGISMOS-based cartilage segmentation, a fully automated meniscus segmentation system was build using pattern recognition technique. The leave-one-out test for the nine training images showed very good mean DSC 0.80+-0.04. The signed and unsigned surface positioning error when compared to manually-traced independent standard in the 60 randomly selected 3D MR image datasets is 0.65+-0.20 and 0.68+-0.20 mm respectively.