Öztürk Ceyda Nur, Albayrak Songül
Yıldız Technical University, Computer Engineering Department, Istanbul, Turkey.
Comput Biol Med. 2016 May 1;72:90-107. doi: 10.1016/j.compbiomed.2016.03.011. Epub 2016 Mar 18.
Anatomical structures that can deteriorate over time, such as cartilage, can be successfully delineated with voxel-classification approaches in magnetic resonance (MR) images. However, segmentation via voxel-classification is a computationally demanding process for high-field MR images with high spatial resolutions. In this study, the whole femoral, tibial, and patellar cartilage compartments in the knee joint were automatically segmented in high-field MR images obtained from Osteoarthritis Initiative using a voxel-classification-driven region-growing algorithm with sample-expand method. Computational complexity of the classification was alleviated via subsampling of the background voxels in the training MR images and selecting a small subset of significant features by taking into consideration systems with limited memory and processing power. Although subsampling of the voxels may lead to a loss of generality of the training models and a decrease in segmentation accuracies, effective subsampling strategies can overcome these problems. Therefore, different subsampling techniques, which involve uniform, Gaussian, vicinity-correlated (VC) sparse, and VC dense subsampling, were used to generate four training models. The segmentation system was experimented using 10 training and 23 testing MR images, and the effects of different training models on segmentation accuracies were investigated. Experimental results showed that the highest mean Dice similarity coefficient (DSC) values for all compartments were obtained when the training models of VC sparse subsampling technique were used. Mean DSC values optimized with this technique were 82.6%, 83.1%, and 72.6% for femoral, tibial, and patellar cartilage compartments, respectively, when mean sensitivities were 79.9%, 84.0%, and 71.5%, and mean specificities were 99.8%, 99.9%, and 99.9%.
随着时间推移会发生退变的解剖结构，如软骨，可通过磁共振（MR）图像中的体素分类方法成功描绘。然而，对于具有高空间分辨率的高场MR图像，通过体素分类进行分割是一个计算量很大的过程。在本研究中，使用基于体素分类驱动的区域生长算法和样本扩展方法，对从骨关节炎倡议组织获得的高场MR图像中的膝关节全股骨、胫骨和髌软骨区域进行自动分割。通过对训练MR图像中的背景体素进行子采样，并考虑到内存和处理能力有限的系统，选择一小部分显著特征，降低了分类的计算复杂度。尽管体素子采样可能会导致训练模型的通用性丧失和分割精度下降，但有效的子采样策略可以克服这些问题。因此，使用了不同的子采样技术，包括均匀、高斯、邻域相关（VC）稀疏和VC密集子采样，以生成四个训练模型。使用10幅训练MR图像和23幅测试MR图像对分割系统进行实验，并研究不同训练模型对分割精度的影响。实验结果表明，使用VC稀疏子采样技术的训练模型时，所有区域的平均骰子相似系数（DSC）值最高。当平均敏感度分别为79.9%、84.0%和71.5%，平均特异性分别为99.8%、99.9%和99.9%时，该技术优化后的股骨、胫骨和髌软骨区域的平均DSC值分别为82.6%、83.1%和72.6%。
