Benchmarking the CoW with the TopCoW Challenge: Topology-Aware Anatomical Segmentation of the Circle of Willis for CTA and MRA

The Circle of Willis (CoW) is an important network of arteries connecting major circulations of the brain. Its vascular architecture is believed to affect the risk, severity, and clinical outcome of serious neuro-vascular diseases. However, characterizing the highly variable CoW anatomy is still a manual and time-consuming expert task. The CoW is usually imaged by two angiographic imaging modalities, magnetic resonance angiography (MRA) and computed tomography angiography (CTA), but there exist limited public datasets with annotations on CoW anatomy, especially for CTA. Therefore we organized the TopCoW Challenge in 2023 with the release of an annotated CoW dataset. The TopCoW dataset was the first public dataset with voxel-level annotations for thirteen possible CoW vessel components, enabled by virtual-reality (VR) technology. It was also the first large dataset with paired MRA and CTA from the same patients. TopCoW challenge formalized the CoW characterization problem as a multiclass anatomical segmentation task with an emphasis on topological metrics. We invited submissions worldwide for the CoW segmentation task, which attracted over 140 registered participants from four continents. The top performing teams managed to segment many CoW components to Dice scores around 90%, but with lower scores for communicating arteries and rare variants. There were also topological mistakes for predictions with high Dice scores. Additional topological analysis revealed further areas for improvement in detecting certain CoW components and matching CoW variant topology accurately. TopCoW represented a first attempt at benchmarking the CoW anatomical segmentation task for MRA and CTA, both morphologically and topologically.

Denoising diffusion-based MRI to CT image translation enables automated spinal segmentation

BACKGROUND

Automated segmentation of spinal magnetic resonance imaging (MRI) plays a vital role both scientifically and clinically. However, accurately delineating posterior spine structures is challenging.

METHODS

This retrospective study, approved by the ethical committee, involved translating T1-weighted and T2-weighted images into computed tomography (CT) images in a total of 263 pairs of CT/MR series. Landmark-based registration was performed to align image pairs. We compared two-dimensional (2D) paired - Pix2Pix, denoising diffusion implicit models (DDIM) image mode, DDIM noise mode - and unpaired (SynDiff, contrastive unpaired translation) image-to-image translation using "peak signal-to-noise ratio" as quality measure. A publicly available segmentation network segmented the synthesized CT datasets, and Dice similarity coefficients (DSC) were evaluated on in-house test sets and the "MRSpineSeg Challenge" volumes. The 2D findings were extended to three-dimensional (3D) Pix2Pix and DDIM.

RESULTS

2D paired methods and SynDiff exhibited similar translation performance and DCS on paired data. DDIM image mode achieved the highest image quality. SynDiff, Pix2Pix, and DDIM image mode demonstrated similar DSC (0.77). For craniocaudal axis rotations, at least two landmarks per vertebra were required for registration. The 3D translation outperformed the 2D approach, resulting in improved DSC (0.80) and anatomically accurate segmentations with higher spatial resolution than that of the original MRI series.

CONCLUSIONS

Two landmarks per vertebra registration enabled paired image-to-image translation from MRI to CT and outperformed all unpaired approaches. The 3D techniques provided anatomically correct segmentations, avoiding underprediction of small structures like the spinous process.

RELEVANCE STATEMENT

This study addresses the unresolved issue of translating spinal MRI to CT, making CT-based tools usable for MRI data. It generates whole spine segmentation, previously unavailable in MRI, a prerequisite for biomechanical modeling and feature extraction for clinical applications.

KEY POINTS

• Unpaired image translation lacks in converting spine MRI to CT effectively. • Paired translation needs registration with two landmarks per vertebra at least. • Paired image-to-image enables segmentation transfer to other domains. • 3D translation enables super resolution from MRI to CT. • 3D translation prevents underprediction of small structures.

Incidental vertebral fracture prediction using neuronal network-based automatic spine segmentation and volumetric bone mineral density extraction from routine clinical CT scans

Objectives

To investigate vertebral osteoporotic fracture (VF) prediction by automatically extracted trabecular volumetric bone mineral density (vBMD) from routine CT, and to compare the model with fracture prevalence-based prediction models.

Methods

This single-center retrospective study included patients who underwent two thoraco-abdominal CT scans during clinical routine with an average inter-scan interval of 21.7 ± 13.1 months (range 5-52 months). Automatic spine segmentation and vBMD extraction was performed by a convolutional neural network framework (anduin.bonescreen.de). Mean vBMD was calculated for levels T5-8, T9-12, and L1-5. VFs were identified by an expert in spine imaging. Odds ratios (ORs) for prevalent and incident VFs were calculated for vBMD (per standard deviation decrease) at each level, for baseline VF prevalence (yes/no), and for baseline VF count (n) using logistic regression models, adjusted for age and sex. Models were compared using Akaike's and Bayesian information criteria (AIC & BIC).

Results

420 patients (mean age, 63 years ± 9, 276 males) were included in this study. 40 (25 female) had prevalent and 24 (13 female) had incident VFs. Individuals with lower vBMD at any spine level had higher odds for VFs (L1-5, prevalent VF: OR,95%-CI,p: 2.2, 1.4-3.5,p=0.001; incident VF: 3.5, 1.8-6.9,p<0.001). In contrast, VF status (2.15, 0.72-6.43,p=0.170) and count (1.38, 0.89-2.12,p=0.147) performed worse in incident VF prediction. Information criteria revealed best fit for vBMD-based models (AIC vBMD=165.2; VF status=181.0; count=180.7).

Conclusions

VF prediction based on automatically extracted vBMD from routine clinical MDCT outperforms prediction models based on VF status and count. These findings underline the importance of opportunistic quantitative osteoporosis screening in clinical routine MDCT data.

The Liver Tumor Segmentation Benchmark (LiTS)

In this work, we report the set-up and results of the Liver Tumor Segmentation Benchmark (LiTS), which was organized in conjunction with the IEEE International Symposium on Biomedical Imaging (ISBI) 2017 and the International Conferences on Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2017 and 2018. The image dataset is diverse and contains primary and secondary tumors with varied sizes and appearances with various lesion-to-background levels (hyper-/hypo-dense), created in collaboration with seven hospitals and research institutions. Seventy-five submitted liver and liver tumor segmentation algorithms were trained on a set of 131 computed tomography (CT) volumes and were tested on 70 unseen test images acquired from different patients. We found that not a single algorithm performed best for both liver and liver tumors in the three events. The best liver segmentation algorithm achieved a Dice score of 0.963, whereas, for tumor segmentation, the best algorithms achieved Dices scores of 0.674 (ISBI 2017), 0.702 (MICCAI 2017), and 0.739 (MICCAI 2018). Retrospectively, we performed additional analysis on liver tumor detection and revealed that not all top-performing segmentation algorithms worked well for tumor detection. The best liver tumor detection method achieved a lesion-wise recall of 0.458 (ISBI 2017), 0.515 (MICCAI 2017), and 0.554 (MICCAI 2018), indicating the need for further research. LiTS remains an active benchmark and resource for research, e.g., contributing the liver-related segmentation tasks in http://medicaldecathlon.com/. In addition, both data and online evaluation are accessible via https://competitions.codalab.org/competitions/17094.

Validation of a Patient-Specific Musculoskeletal Model for Lumbar Load Estimation Generated by an Automated Pipeline From Whole Body CT

Background: Chronic back pain is a major health problem worldwide. Although its causes can be diverse, biomechanical factors leading to spinal degeneration are considered a central issue. Numerical biomechanical models can identify critical factors and, thus, help predict impending spinal degeneration. However, spinal biomechanics are subject to significant interindividual variations. Therefore, in order to achieve meaningful findings on potential pathologies, predictive models have to take into account individual characteristics. To make these highly individualized models suitable for systematic studies on spinal biomechanics and clinical practice, the automation of data processing and modeling itself is inevitable. The purpose of this study was to validate an automatically generated patient-specific musculoskeletal model of the spine simulating static loading tasks.

Methods: CT imaging data from two patients with non-degenerative spines were processed using an automated deep learning-based segmentation pipeline. In a semi-automated process with minimal user interaction, we generated patient-specific musculoskeletal models and simulated various static loading tasks. To validate the model, calculated vertebral loadings of the lumbar spine and muscle forces were compared with in vivo data from the literature. Finally, results from both models were compared to assess the potential of our process for interindividual analysis.

Results: Calculated vertebral loads and muscle activation overall stood in close correlation with data from the literature. Compression forces normalized to upright standing deviated by a maximum of 16% for flexion and 33% for lifting tasks. Interindividual comparison of compression, as well as lateral and anterior–posterior shear forces, could be linked plausibly to individual spinal alignment and bodyweight.

Conclusion: We developed a method to generate patient-specific musculoskeletal models of the lumbar spine. The models were able to calculate loads of the lumbar spine for static activities with respect to individual biomechanical properties, such as spinal alignment, bodyweight distribution, and ligament and muscle insertion points. The process is automated to a large extent, which makes it suitable for systematic investigation of spinal biomechanics in large datasets.

A Unified 3D Framework for Organs-at-Risk Localization and Segmentation for Radiation Therapy Planning

Automatic localization and segmentation of organs-at-risk (OAR) in CT are essential pre-processing steps in medical image analysis tasks, such as radiation therapy planning. For instance, the segmentation of OAR surrounding tumors enables the maximization of radiation to the tumor area without compromising the healthy tissues. However, the current medical workflow requires manual delineation of OAR, which is prone to errors and is annotator-dependent. In this work, we aim to introduce a unified 3D pipeline for OAR localization-segmentation rather than novel localization or segmentation architectures. To the best of our knowledge, our proposed framework fully enables the exploitation of 3D context information inherent in medical imaging. In the first step, a 3D multi-variate regression network predicts organs' centroids and bounding boxes. Secondly, 3D organ-specific segmentation networks are leveraged to generate a multi-organ segmentation map. Our method achieved an overall Dice score of 0.9260 ± 0.18% on the VISCERAL dataset containing CT scans with varying fields of view and multiple organs.

Self-Supervised Pretext Tasks in Model Robustness & Generalizability: A Revisit from Medical Imaging Perspective

Self-supervised pretext tasks have been introduced as an effective strategy when learning target tasks on small annotated data sets. However, while current research focuses on exploring novel pretext tasks for meaningful and reusable representation learning for the target task, the study of its robustness and generalizability has remained relatively under-explored. Specifically, it is crucial in medical imaging to proactively investigate performance under different perturbations for reliable deployment of clinical applications. In this work, we revisit medical imaging networks pre-trained with self-supervised learnings and categorically evaluate robustness and generalizability compared to vanilla supervised learning. Our experiments on pneumonia detection in X-rays and multi-organ segmentation in CT yield conclusive results exposing the hidden benefits of self-supervision pre-training for learning robust feature representations.

Automated Opportunistic Osteoporosis Screening in Routine Computed Tomography of the Spine: Comparison With Dedicated Quantitative CT

Opportunistic osteoporosis screening in nondedicated routine computed tomography (CT) is of increasing importance. The purpose of this study was to compare lumbar volumetric bone mineral density (vBMD) assessed by a convolutional neural network (CNN)-based framework in routine CT to vBMD from dedicated quantitative CT (QCT), and to evaluate the ability of vBMD and surrogate measurements of Hounsfield units (HU) to distinguish between patients with and without osteoporotic vertebral fractures (VFs). A total of 144 patients (median age: 70.7 years, 93 females) with clinical routine CT (eight different CT scanners, 120 kVp or 140 kVp, with and without intravenous contrast medium) and dedicated QCT acquired within ≤30 days were included. Vertebral measurements included (i) vBMD from the CNN-based approach including automated vertebral body labeling, segmentation, and correction of the contrast media phase for routine CT data (vBMD_OPP), (ii) vBMD from dedicated QCT (vBMD_QCT), and (iii) noncalibrated HU from vertebral bodies of routine CT data as previously proposed for immanent opportunistic osteoporosis screening based on CT attenuation. The intraclass correlation coefficient (ICC) for vBMD_QCT versus vBMD_OPP indicated better agreement (ICC = 0.913) than the ICC for vBMD_QCT versus noncalibrated HU (ICC = 0.704). Bland-Altman analysis showed data points from 137 patients (95.1%) within the limits of agreement (LOA) of -23.2 to 25.0 mg/cm³ for vBMD_QCT versus vBMD_OPP. Osteoporosis (vBMD <80 mg/cm³ ) was detected in 89 patients (vBMD_QCT) and 88 patients (vBMD_OPP), whereas no patient crossed the diagnostic thresholds from normal vBMD to osteoporosis or vice versa. In a subcohort of 88 patients (thoracolumbar spine covered by imaging for VF reading), 69 patients showed one or more prevalent VFs, and the performance for discrimination between patients with and without VFs was best for vBMD_OPP (area under the curve [AUC] = 0.862; 95% confidence interval [CI], 0.771-0.953). In conclusion, automated opportunistic osteoporosis screening in routine CT of various scanner setups is feasible and may demonstrate high diagnostic accuracy for prevalent VFs. © 2022 The Authors. Journal of Bone and Mineral Research published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research (ASBMR).

Anatomy-Aware Inference of the 3D Standing Spine Posture from 2D Radiographs

An important factor for the development of spinal degeneration, pain and the outcome of spinal surgery is known to be the balance of the spine. It must be analyzed in an upright, standing position to ensure physiological loading conditions and visualize load-dependent deformations. Despite the complex 3D shape of the spine, this analysis is currently performed using 2D radiographs, as all frequently used 3D imaging techniques require the patient to be scanned in a prone position. To overcome this limitation, we propose a deep neural network to reconstruct the 3D spinal pose in an upright standing position, loaded naturally. Specifically, we propose a novel neural network architecture, which takes orthogonal 2D radiographs and infers the spine’s 3D posture using vertebral shape priors. In this work, we define vertebral shape priors using an atlas and a spine shape prior, incorporating both into our proposed network architecture. We validate our architecture on digitally reconstructed radiographs, achieving a 3D reconstruction Dice of 0.95, indicating an almost perfect 2D-to-3D domain translation. Validating the reconstruction accuracy of a 3D standing spine on real data is infeasible due to the lack of a valid ground truth. Hence, we design a novel experiment for this purpose, using an orientation invariant distance metric, to evaluate our model’s ability to synthesize full-3D, upright, and patient-specific spine models. We compare the synthesized spine shapes from clinical upright standing radiographs to the same patient’s 3D spinal posture in the prone position from CT.

Level-Specific Volumetric BMD Threshold Values for the Prediction of Incident Vertebral Fractures Using Opportunistic QCT: A Case-Control Study

PURPOSE

To establish and evaluate the diagnostic accuracy of volumetric bone mineral density (vBMD) threshold values at different spinal levels, derived from opportunistic quantitative computed tomography (QCT), for the prediction of incident vertebral fractures (VF).

MATERIALS AND METHODS

In this case-control study, 35 incident VF cases (23 women, 12 men; mean age: 67 years) and 70 sex- and age-matched controls were included, based on routine multi detector CT (MDCT) scans of the thoracolumbar spine. Trabecular vBMD was measured from routine baseline CT scans of the thoracolumbar spine using an automated pipeline including vertebral segmentation, asynchronous calibration for HU-to-vBMD conversion, and correction of intravenous contrast medium (https://anduin.bonescreen.de). Threshold values at T1-L5 were calculated for the optimal operating point according to the Youden index and for fixed sensitivities (60 - 85%) in receiver operating characteristic (ROC) curves.

RESULTS

vBMD at each single level of the thoracolumbar spine was significantly associated with incident VFs (odds ratio per SD decrease [OR], 95% confidence interval [CI] at T1-T4: 3.28, 1.66-6.49; at T5-T8: 3.28, 1.72-6.26; at T9-T12: 3.37, 1.78-6.36; and at L1-L4: 3.98, 1.97-8.06), independent of adjustment for age, sex, and prevalent VF. AUC showed no significant difference between vertebral levels and was highest at the thoracolumbar junction (AUC = 0.75, 95%-CI = 0.63 - 0.85 for T11-L2). Optimal threshold values increased from lumbar (L1-L4: 52.0 mg/cm³) to upper thoracic spine (T1-T4: 69.3 mg/cm³). At T11-L2, T12-L3 and L1-L4, a threshold of 80.0 mg/cm³ showed sensitivities of 85 - 88%, and specificities of 41 - 49%. To achieve comparable sensitivity (85%) at more superior spinal levels, resulting thresholds were higher: 114.1 mg/cm³ (T1-T4), 92.0 mg/cm³ (T5-T8), 88.2 mg/cm³ (T9-T12).

CONCLUSIONS

At all levels of the thoracolumbar spine, lower vBMD was associated with incident VFs in an elderly, predominantly oncologic patient population. Automated opportunistic osteoporosis screening of vBMD along the entire thoracolumbar spine allows for risk assessment of imminent VFs. We propose level-specific vBMD threshold at the thoracolumbar spine to identify individuals at high fracture risk.

Automatic opportunistic osteoporosis screening in routine CT: improved prediction of patients with prevalent vertebral fractures compared to DXA

OBJECTIVES

To compare spinal bone measures derived from automatic and manual assessment in routine CT with dual energy X-ray absorptiometry (DXA) in their association with prevalent osteoporotic vertebral fractures using our fully automated framework ( https://anduin.bonescreen.de ) to assess various bone measures in clinical CT.

METHODS

We included 192 patients (141 women, 51 men; age 70.2 ± 9.7 years) who had lumbar DXA and CT available (within 1 year). Automatic assessment of spinal bone measures in CT included segmentation of vertebrae using a convolutional neural network (CNN), reduction to the vertebral body, and extraction of bone mineral content (BMC), trabecular and integral volumetric bone mineral density (vBMD), and CT-based areal BMD (aBMD) using asynchronous calibration. Moreover, trabecular bone was manually sampled (manual vBMD).

RESULTS

A total of 148 patients (77%) had vertebral fractures and significantly lower values in all bone measures compared to patients without fractures (p ≤ 0.001). Except for BMC, all CT-based measures performed significantly better as predictors for vertebral fractures compared to DXA (e.g., AUC = 0.885 for trabecular vBMD and AUC = 0.86 for integral vBMD vs. AUC = 0.668 for DXA aBMD, respectively; both p < 0.001). Age- and sex-adjusted associations with fracture status were strongest for manual vBMD (OR = 7.3, [95%] CI 3.8-14.3) followed by automatically assessed trabecular vBMD (OR = 6.9, CI 3.5-13.4) and integral vBMD (OR = 4.3, CI 2.5-7.6). Diagnostic cutoffs of integral vBMD for osteoporosis (< 160 mg/cm3) or low bone mass (160 ≤ BMD < 190 mg/cm3) had sensitivity (84%/41%) and specificity (78%/95%) similar to trabecular vBMD.

CONCLUSIONS

Fully automatic osteoporosis screening in routine CT of the spine is feasible. CT-based measures can better identify individuals with reduced bone mass who suffered from vertebral fractures than DXA.

KEY POINTS

• Opportunistic osteoporosis screening of spinal bone measures derived from clinical routine CT is feasible in a fully automatic fashion using a deep learning-driven framework ( https://anduin.bonescreen.de ). • Manually sampled volumetric BMD (vBMD) and automatically assessed trabecular and integral vBMD were the best predictors for prevalent vertebral fractures. • Except for bone mineral content, all CT-based bone measures performed significantly better than DXA-based measures. • We introduce diagnostic thresholds of integral vBMD for osteoporosis (< 160 mg/cm3) and low bone mass (160 ≤ BMD < 190 mg/cm3) with almost equal sensitivity and specificity compared to conventional thresholds of quantitative CT as proposed by the American College of Radiology (osteoporosis < 80 mg/cm3).

VerSe: A Vertebrae labelling and segmentation benchmark for multi-detector CT images

Vertebral labelling and segmentation are two fundamental tasks in an automated spine processing pipeline. Reliable and accurate processing of spine images is expected to benefit clinical decision support systems for diagnosis, surgery planning, and population-based analysis of spine and bone health. However, designing automated algorithms for spine processing is challenging predominantly due to considerable variations in anatomy and acquisition protocols and due to a severe shortage of publicly available data. Addressing these limitations, the Large Scale Vertebrae Segmentation Challenge (VerSe) was organised in conjunction with the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in 2019 and 2020, with a call for algorithms tackling the labelling and segmentation of vertebrae. Two datasets containing a total of 374 multi-detector CT scans from 355 patients were prepared and 4505 vertebrae have individually been annotated at voxel level by a human-machine hybrid algorithm (https://osf.io/nqjyw/, https://osf.io/t98fz/). A total of 25 algorithms were benchmarked on these datasets. In this work, we present the results of this evaluation and further investigate the performance variation at the vertebra level, scan level, and different fields of view. We also evaluate the generalisability of the approaches to an implicit domain shift in data by evaluating the top-performing algorithms of one challenge iteration on data from the other iteration. The principal takeaway from VerSe: the performance of an algorithm in labelling and segmenting a spine scan hinges on its ability to correctly identify vertebrae in cases of rare anatomical variations. The VerSe content and code can be accessed at: https://github.com/anjany/verse.

Prediction of incident vertebral fractures in routine MDCT: Comparison of global texture features, 3D finite element parameters and volumetric BMD

PURPOSE

In this case-control study, we evaluated different quantitative parameters derived from routine multi-detector computed tomography (MDCT) scans with respect to their ability to predict incident osteoporotic vertebral fractures of the thoracolumbar spine.

METHODS

16 patients who received baseline and follow-up contrast-enhanced MDCT and were diagnosed with an incident osteoporotic vertebral fracture at follow-up, and 16 age-, sex-, and follow-up-time-matched controls were included in the study. Vertebrae were labelled and segmented using a fully automated pipeline. Volumetric bone mineral density (vBMD), finite element analysis (FEA)-based failure load (FL) and failure displacement (FD), as well as 24 texture features were extracted from L1 - L3 and averaged. Odds ratios (OR) with 95% confidence intervals (CI), expressed per standard deviation decrease, receiver operating characteristic (ROC) area under the curve (AUC), as well as logistic regression models, including all analyzed parameters as independent variables, were used to assess the prediction of incident vertebral fractures.

RESULTS

The texture feature Correlation (AUC = 0.754, p = 0.014; OR = 2.76, CI = 1.16-6.58) and vBMD (AUC = 0.750, p = 0.016; OR = 2.67, CI = 1.12-6.37) classified incident vertebral fractures best, while the best FEA-based parameter FL showed an AUC = 0.719 (p = 0.035). Correlation was the only significant predictor of incident fractures in the logistic regression analysis of all parameters (p = 0.022).

CONCLUSION

MDCT-derived FEA parameters and texture features, averaged from L1 - L3, showed only a moderate, but no statistically significant improvement of incident vertebral fracture prediction beyond BMD, supporting the hypothesis that vertebral-specific parameters may be superior for fracture risk assessment.

MDCT-Based Finite Element Analyses: Are Measurements at the Lumbar Spine Associated with the Biomechanical Strength of Functional Spinal Units of Incidental Osteoporotic Fractures along the Thoracolumbar Spine?

Assessment of osteoporosis-associated fracture risk during clinical routine is based on the evaluation of clinical risk factors and T-scores, as derived from measurements of areal bone mineral density (aBMD). However, these parameters are limited in their ability to identify patients at high fracture risk. Finite element models (FEMs) have shown to improve bone strength prediction beyond aBMD. This study aims to investigate whether FEM measurements at the lumbar spine can predict the biomechanical strength of functional spinal units (FSUs) with incidental osteoporotic vertebral fractures (VFs) along the thoracolumbar spine. Multi-detector computed tomography (MDCT) data of 11 patients (5 females and 6 males, median age: 67 years) who underwent MDCT twice (median interval between baseline and follow-up MDCT: 18 months) and sustained an incidental osteoporotic VF between baseline and follow-up scanning were used. Based on baseline MDCT data, two FSUs consisting of vertebral bodies and intervertebral discs (IVDs) were modeled: one standardly capturing L1-IVD-L2-IVD-L3 (FSU_L1-L3) and one modeling the incidentally fractured vertebral body at the center of the FSU (FSU_F). Furthermore, volumetric BMD (vBMD) derived from MDCT, FEM-based displacement, and FEM-based load of the single vertebrae L1 to L3 were determined. Statistically significant correlations (adjusted for a BMD ratio of fracture/L1-L3 segments) were revealed between the FSU_F and mean load of L1 to L3 (r = 0.814, p = 0.004) and the mean vBMD of L1 to L3 (r = 0.745, p = 0.013), whereas there was no statistically significant association between the FSU_F and FSU_L1-L3 or between FSU_F and the mean displacement of L1 to L3 (p > 0.05). In conclusion, FEM measurements of single vertebrae at the lumbar spine may be able to predict the biomechanical strength of incidentally fractured vertebral segments along the thoracolumbar spine, while FSUs seem to predict only segment-specific fracture risk.

Prediction of Incidental Osteoporotic Fractures at Vertebral-Specific Level Using 3D Non-Linear Finite Element Parameters Derived from Routine Abdominal MDCT

To investigate whether finite element (FE) analysis of the spine in routine thoracic/abdominal multi-detector computed tomography (MDCT) can predict incidental osteoporotic fractures at vertebral-specific level; Baseline routine thoracic/abdominal MDCT scans of 16 subjects (8(m), mean age: 66.1 ± 8.2 years and 8(f), mean age: 64.3 ± 9.5 years) who sustained incidental osteoporotic vertebral fractures as confirmed in follow-up MDCTs were included in the current study. Thoracic and lumbar vertebrae (T5-L5) were automatically segmented, and bone mineral density (BMD), finite element (FE)-based failure-load, and failure-displacement were determined. These values of individual vertebrae were normalized globally (g), by dividing the absolute value with the average of L1-3 and locally by dividing the absolute value with the average of T5-12 and L1-5 for thoracic and lumbar vertebrae, respectively. Mean-BMD of L1-3 was determined as reference. Receiver operating characteristics (ROC) and area under the curve (AUC) were calculated for different normalized FE (Kload, Kdisplacement,K(load)g, and K(displacement)g) and BMD (KBMD, and K(BMD)g) ratio parameter combinations for identifying incidental fractures. Kload, K(load)g, KBMD, and K(BMD)g showed significantly higher discriminative power compared to standard mean BMD of L1-3 (BMDStandard) (AUC = 0.67 for Kload; 0.64 for K(load)g; 0.64 for KBMD; 0.61 for K(BMD)g vs. 0.54 for BMDStandard). The combination of Kload, Kdisplacement, and KBMD increased the AUC further up to 0.77 (p < 0.001). The combination of FE with BMD measurements derived from routine thoracic/abdominal MDCT allowed an improved prediction of incidental fractures at vertebral-specific level.

Gender-, Age- and Region-Specific Characterization of Vertebral Bone Microstructure Through Automated Segmentation and 3D Texture Analysis of Routine Abdominal CT

PURPOSE

To identify long-term reproducible texture features (TFs) of spinal computed tomography (CT), and characterize variations with regard to gender, age and vertebral level using our automated quantification framework.

METHODS

We performed texture analysis (TA) on baseline and follow-up CT (follow-up duration: 30-90 days) of 21 subjects (8 females, 13 males, age at baseline 61.2 ± 9.2 years) to determine long-term reproducibility. TFs with a long-term reproducibility error Δ_rel<5% were further analyzed for an association with age and vertebral level in a cohort of 376 patients (129 females, 247 males, age 62.5 ± 9.2 years). Automated analysis comprised labeling and segmentation of vertebrae into subregions using a convolutional neural network, calculation of volumetric bone mineral density (vBMD) with asynchronous calibration and TF extraction. Variance_global measures the spread of the gray-level distribution in an image while Entropy reflects the uniformity of gray-levels. Short-run emphasis (SRE), Long-run emphasis (LRE), Run-length non-uniformity (RLN) and Run percentage (RP) contain information on consecutive voxels of a particular grey-level, or grey-level range, in a particular direction. Long runs (LRE) represent coarse texture while short runs (SRE) represent fine texture. RLN reflects similarities in the length of runs while RP reflects distribution and homogeneity of runs with a specific direction.

RESULTS

Six of the 24 extracted TFs had Δ_rel<5% (Variance_global, Entropy, SRE, LRE, RLN, RP), and were analyzed further in 4716 thoracolumbar vertebrae. Five TFs (Variance_global,SRE,LRE, RLN,RP) showed a significant difference between genders (p<0.001), potentially being caused by a finer and more directional vertebral trabecular microstructure in females compared to males. Variance_global and Entropy showed a significant increase from the thoracic to the lumbar spine (p<0.001), indicating a higher degree and earlier initiation of trabecular microstructure deterioration at lower spinal levels. The four higher-order TFs showed significant variations between spine regions without a clear directional gradient (p ≤ 0.001-0.012). No TF showed a clear age dependence. vBMD differed significantly between genders, age groups and spine regions (p ≤ 0.001-0.002).

CONCLUSION

Long-term reproducible CT-based TFs of the thoracolumbar spine were established and characterized in a predominantly older adult study population. The gender-, age- and vertebral-level-specific values may serve as foundation for osteoporosis diagnostics and facilitate future studies investigating vertebral microstructure.

A Vertebral Segmentation Dataset with Fracture Grading

Published under a CC BY 4.0 license. Supplemental material is available for this article.

Labeling Vertebrae with Two-dimensional Reformations of Multidetector CT Images: An Adversarial Approach for Incorporating Prior Knowledge of Spine Anatomy

Purpose

To use and test a labeling algorithm that operates on two-dimensional reformations, rather than three-dimensional data to locate and identify vertebrae.

Materials and Methods

The authors improved the Btrfly Net, a fully convolutional network architecture described by Sekuboyina et al, which works on sagittal and coronal maximum intensity projections (MIPs) and augmented it with two additional components: spine localization and adversarial a priori learning. Furthermore, two variants of adversarial training schemes that incorporated the anatomic a priori knowledge into the Btrfly Net were explored. The superiority of the proposed approach for labeling vertebrae on three datasets was investigated: a public benchmarking dataset of 302 CT scans and two in-house datasets with a total of 238 CT scans. The Wilcoxon signed rank test was employed to compute the statistical significance of the improvement in performance observed with various architectural components in the authors' approach.

Results

On the public dataset, the authors' approach using the described Btrfly Net with energy-based prior encoding (Btrfly_pe-eb) network performed as well as current state-of-the-art methods, achieving a statistically significant (P < .001) vertebrae identification rate of 88.5% ± 0.2 (standard deviation) and localization distances of less than 7 mm. On the in-house datasets that had a higher interscan data variability, an identification rate of 85.1% ± 1.2 was obtained.

Conclusion

An identification performance comparable to existing three-dimensional approaches was achieved when labeling vertebrae on two-dimensional MIPs. The performance was further improved using the proposed adversarial training regimen that effectively enforced local spine a priori knowledge during training. Spine localization increased the generalizability of our approach by homogenizing the content in the MIPs.Supplemental material is available for this article.© RSNA, 2020.