Spatial normalization of spine MR images for statistical correlation of lesions with clinical symptoms

Fully automated segmentation of lumbar bone marrow in sagittal, high-resolution T1-weighted magnetic resonance images using 2D U-NET

BACKGROUND

We investigated a 2-dimensional (2D) U-Net model to delineate lumbar bone marrow (BM) using a high resolution T1-weighted magnetic resonance imaging.

METHOD

Healthy controls (n = 44, 836 images) and patients with hematologic diseases (n = 56, 1064 images) received MRI of the lumbar spines. Lumbar BM on each image was manually delineated by an experienced radiologist as a ground-truth. The 2D U-Net models were trained using a healthy lumbar BM only, diseased BM only, and using healthy and diseased BM combined, respectively. The models were validated using healthy and diseased subjects, separately. A repeated-measures analysis of variance was performed to compare segmentation accuracies with 2 validation cohorts among U-Net trained with healthy subjects (UNET_HC), U-Net trained with diseased subjects (UNET_HD), U-Net trained with all subjects including both healthy and diseased subjects (UNET_HCHD), and 3-dimensional Grow-Cut algorithm (3DGC).

RESULTS

When validated with the healthy subjects, UNET_HC, UNET_HD, UNET_HCHD and 3DGC achieved the mean and standard deviation of the Dice Similarity Coefficient (DSC) of 0.9415 ± 0.07056, 0.9583 ± 0.05146, 0.9602 ± 0.0486 and 0.9139 ± 0.2039, respectively. When validated with the diseased subjects, DSCs of UNET_HC, UNET_HD, UNET_HCHD and 3DGC were 0.8303 ± 0.1073, 0.9502 ± 0.0217, 0.9502 ± 0.0217 and 0.8886 ± 0.2179, respectively. The U-Net models segmented BM better than the semi-automatic 3DGC (P < 0.0001), and UNET_HD produced better results than UNET_HC (P < 0.0001).

CONCLUSIONS

We successfully constructed a fully automatic lumbar BM segmentation model for a high-resolution T1-weighted MRI using U-Net, which outperformed most of the previously reported approaches and the existing semi-automatic algorithm.

Automatic detection and segmentation of lumbar vertebrae from X-ray images for compression fracture evaluation

For compression fracture detection and evaluation, an automatic X-ray image segmentation technique that combines deep-learning and level-set methods is proposed. Automatic segmentation is much more difficult for X-ray images than for CT or MRI images because they contain overlapping shadows of thoracoabdominal structures including lungs, bowel gases, and other bony structures such as ribs. Additional difficulties include unclear object boundaries, the complex shape of the vertebra, inter-patient variability, and variations in image contrast. Accordingly, a structured hierarchical segmentation method is presented that combines the advantages of two deep-learning methods. Pose-driven learning is used to selectively identify the five lumbar vertebrae in an accurate and robust manner. With knowledge of the vertebral positions, M-net is employed to segment the individual vertebra. Finally, fine-tuning segmentation is applied by combining the level-set method with the previously obtained segmentation results. The performance of the proposed method was validated by 160 lumbar X-ray images, resulting in a mean Dice similarity metric of 91.60±2.22%. The results show that the proposed method achieves accurate and robust identification of each lumbar vertebra and fine segmentation of individual vertebra.

Vertebral body segmentation in wide range clinical routine spine MRI data

BACKGROUND AND OBJECTIVE

In this work we propose a 3D vertebral body segmentation approach for clinical magnetic resonance (MR) spine imaging. So far, vertebrae segmentation approaches in MR spine imaging are either limited to particular MR imaging sequences or require minutes to compute, which can be hindering in clinical routine. The major contribution of our work is a reasonably precise segmentation result, within seconds and with minimal user interaction, for spine MR imaging commonly used in clinical routine. Our focus lies on the applicability towards a large variety of clinical MR imaging sequences, dealing with low image quality, high anisotropy and spine pathologies.

METHODS

Our method starts with a intensity correction step to deal with bias field artifacts and a minimal user-assisted initialization. Next, appearance-based vertebral body probability maps guide a subsequent hybrid level-set segmentation.

RESULTS

We tested our method on different MR imaging sequences from 48 subjects. Overall, our evaluation set contains 63 datasets including 419 vertebral bodies, which differ in age, sex and presence of spine pathologies. This is the largest set of reference segmentations of clinical routine spine MR imaging so far. We achieved a Dice coefficient of 86.0%, a mean Euclidean surface distance error of 1.59 ± 0.24 mm and a Hausdorff distance of 6.86 mm.

CONCLUSIONS

These results illustrate the robustness of our segmentation approach towards the variety of MR image data, which is a pivotal aspect for clinical usefulness and reliable diagnosis.

Statistical shape model reconstruction with sparse anomalous deformations: Application to intervertebral disc herniation

Many medical image processing techniques rely on accurate shape modeling of anatomical features. The presence of shape abnormalities challenges traditional processing algorithms based on strong morphological priors. In this work, a sparse shape reconstruction from a statistical shape model is presented. It combines the advantages of traditional statistical shape models (defining a 'normal' shape space) and previously presented sparse shape composition (providing localized descriptors of anomalies). The algorithm was incorporated into our image segmentation and classification software. Evaluation was performed on simulated and clinical MRI data from 22 sciatica patients with intervertebral disc herniation, containing 35 herniated and 97 normal discs. Moderate to high correlation (R=0.73) was achieved between simulated and detected herniations. The sparse reconstruction provided novel quantitative features describing the herniation morphology and MRI signal appearance in three dimensions (3D). The proposed descriptors of local disc morphology resulted to the 3D segmentation accuracy of 1.07±1.00mm (mean absolute vertex-to-vertex mesh distance over the posterior disc region), and improved the intervertebral disc classification from 0.888 to 0.931 (area under receiver operating curve). The results show that the sparse shape reconstruction may improve computer-aided diagnosis of pathological conditions presenting local morphological alterations, as seen in intervertebral disc herniation.

Automated detection, 3D segmentation and analysis of high resolution spine MR images using statistical shape models

Recent advances in high resolution magnetic resonance (MR) imaging of the spine provide a basis for the automated assessment of intervertebral disc (IVD) and vertebral body (VB) anatomy. High resolution three-dimensional (3D) morphological information contained in these images may be useful for early detection and monitoring of common spine disorders, such as disc degeneration. This work proposes an automated approach to extract the 3D segmentations of lumbar and thoracic IVDs and VBs from MR images using statistical shape analysis and registration of grey level intensity profiles. The algorithm was validated on a dataset of volumetric scans of the thoracolumbar spine of asymptomatic volunteers obtained on a 3T scanner using the relatively new 3D T2-weighted SPACE pulse sequence. Manual segmentations and expert radiological findings of early signs of disc degeneration were used in the validation. There was good agreement between manual and automated segmentation of the IVD and VB volumes with the mean Dice scores of 0.89 ± 0.04 and 0.91 ± 0.02 and mean absolute surface distances of 0.55 ± 0.18 mm and 0.67 ± 0.17 mm respectively. The method compares favourably to existing 3D MR segmentation techniques for VBs. This is the first time IVDs have been automatically segmented from 3D volumetric scans and shape parameters obtained were used in preliminary analyses to accurately classify (100% sensitivity, 98.3% specificity) disc abnormalities associated with early degenerative changes.