Evaluation and comparison of 3D intervertebral disc localization and segmentation methods for 3D T2 MR data: A grand challenge

Abstract

Highlights• Establish a standard framework with 25 manually annotated 3D T2MRI data for an objective comparison of intervertebral disc (IVD) localization and segmentation methods• Investigate strengths and limitations of a representative selection of the state-of-the-art IVD localization and segmentation methods with a challenge setup• Results achieved by the best algorithms in this study set new frontiers for IVD localization and segmentation from MR data Experimental results show that overall the best localization method achieves a mean localization distance of 0.8mm and the best segmentation method achieves a mean Dice of 91.8%, a mean average absolute distance of 1.1mm and a mean Hausdorff distance of 4.3mm, respectively. The strengths and drawbacks of each method are discussed, which provides insights into the performance of different IVD localization and segmentation methods.

Holistic decomposition convolution for effective semantic segmentation of medical volume images

“…In this section, we present experimental results of the proposed pipeline for volumetric image segmentation. Two datasets, i.e., an in-house dataset consisting of 25 T1 hip MR images with limited field of view and a publicly available dataset from the MICCAI 2015 IVD localization and segmentation challenge [14], are used in our study. More specifically, first, we conduct an ablation study on the in-house hip dataset to evaluate the influence of the shuffling factors and of the underlying FCNs on the performance of the proposed pipeline.…”

“…Based on the findings from the ablation study, we choose the 3D LP-U-net for our remaining studies. Following [14], we used Dice Overlap Coefficients (DOC), Average Surface Distance (ASD) and Hausdorff Distance (HD) as the evaluation metrics.…”

“…As classifying each pixel/voxel in a sliding window fashion results in orders of magnitude of redundant calculation, most of recent works [10,3] are based on Fully Convolutional Networks (FCNs) [7], which can process the input image in an end-to-end way and can provide a full resolution segmentation map [6]. In several biomedical image segmentation benchmarking competitions, methods built on CNNs [10,14] are on the top list of the associated leaderboard. Despite the fact that CNNs-based methods have achieved state-of-the-art performance in many different 2D medical image analysis tasks, in clinical practice, however, a large part of the medical imaging data available is in 3D.…”

Automatic Spine Tissue Segmentation from MRI Data Based on Cascade of Boosted Classifiers and Active Appearance Model

“…Due to the nature of Catmull–Rom splines [28] further optimization can be done to achieve better interpolation results. The paper [62] presents multiple recent methods for intervertebral disc segmentation, which can be treated as a similar task, including Machine Learning and deep learning based approaches. The segmentation results presented in the paper [62] were measured with dice overlap coefficients and varied from 81.6% to 92% for different methods.…”

“…The paper [62] presents multiple recent methods for intervertebral disc segmentation, which can be treated as a similar task, including Machine Learning and deep learning based approaches. The segmentation results presented in the paper [62] were measured with dice overlap coefficients and varied from 81.6% to 92% for different methods. Comparing those results with obtained segmentation and generalization results of 90.19% and 91.37%, one can conclude that presented AAM approach provides a good segmentation performance and moreover can be applied for intervertebral discs localization and segmentation.…”