OP0062 PREDICTIVE VALUE OF BONE TEXTURE FEATURES EXTRACTED BY DEEP LEARNING MODELS FOR THE DETECTION OF OSTEOARTHRITIS: DATA FROM THE OSTEOARTHRITIS INITIATIVE

Background:

Osteoarthritis is a degenerative disorder characterized by radiographic features of asymmetric loss of joint space, subchondral sclerosis, and osteophyte formation. Conventional plain films are essential to detect structural changes in osteoarthritis. Recent evidence suggests that fractal- and entropy-based bone texture parameters may improve the prediction of radiographic osteoarthritis.1In contrast to the fixed texture features, deep learning models allow the comprehensive texture feature extraction and recognition relevant to osteoarthritis.

Objectives:

To assess the predictive value of deep learning-extracted bone texture features in the detection of radiographic osteoarthritis.

Methods:

We used data from the Osteoarthritis Initiative, which is a longitudinal study with 4,796 patients followed up and assessed for osteoarthritis. We used a training set of 25,978 images from 3,086 patients to develop the textual model. We use the BoneFinder software2to do the segmentation of distal femur and proximal tibia. We used the Deep Texture Encoding Network (Deep-TEN)3to encode the bone texture features into a vector, which is fed to a 5-way linear classifier for Kellgren and Lawrence grading for osteoarthritis classification. We also developed a Residual Network with 18 layers (ResNet18) for comparison since it deals with contours as well. Spearman’s correlation coefficient was used to assess the correlation between predicted and reference KL grades. We also test the performance of the model to identify osteoarthritis (KL grade≥2).

Results:

We obtained 6,490 knee radiographs from 446 female and 326 male patients who were not in the training sets to validate the performance of the models. The distribution of the KL grades in the training and testing sets were shown in Table 1. The Spearman’s correlation coefficient was 0.60 for the Deep-TEN and 0.67 for the ResNet18 model. Table 2 shows the performance of the models to detect osteoarthritis. The positive predictive value for Deep-TEN and ResNet18 model classification for OA was 81.37% and 87.46%, respectively.

Table 1

Distribution of KL grades in the training and testing sets.

KL grades01234TotalTraining set1089341.9%458218.7%611423.5%332012.8%7993.1%25,978Testing set247238.1%135320.8%169626.1%77511.9%1943.0%6,490Table 2

Performance matrices of the Deep-Ten and ResNet18 models to detect osteoarthritis

Deep-TENResNet18Sensitivity62.29%(95% CI, 60.42%–64.13%)59.14%(95% CI, 57.24%–61.01%)Specificity90.07%(95% CI, 89.07%–91.00%)94.09%(95% CI, 93.30%–94.82%)Positive predictive value81.37%(95% CI, 79.81%–82.84%)87.46%(95% CI, 85.96%–88.82%)Negative predictive value77.42%(95% CI, 77.64%–79.65%)76.77%(95% CI, 75.93%–77.59%)

Conclusion:

This study demonstrates that the bone texture model performs reasonably well to detect radiographic osteoarthritis with a similar performance to the bone contour model.

References:

[1]Bertalan Z, Ljuhar R, Norman B, et al. Combining fractal- and entropy-based bone texture analysis for the prediction of osteoarthritis: data from the multicenter osteoarthritis study (MOST). Osteoarthritis Cartilage 2018;26:S49.

[2]Lindner C, Wang CW, Huang CT, et al. Fully Automatic System for Accurate Localisation and Analysis of Cephalometric Landmarks in Lateral Cephalograms. Sci Rep 2016;6:33581.

[3]Zhang H, Xue J, Dana K. Deep TEN: Texture Encoding Network. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017:708-17.

Disclosure of Interests:

None declared

GFR Estimation Using a Panel of Filtration Markers in Shanghai and Beijing

RATIONALE & OBJECTIVES

Estimated glomerular filtration rate (eGFR) using creatinine and cystatin C (eGFRcr-cys) may be less accurate compared to measured GFR (mGFR) in China than in North America, Europe, and Australia due to variation across regions in their non-GFR determinants. The non-GFR determinants of β2-microglobulin (B2M) and β-trace protein (BTP) differ from those of creatinine and cystatin C. Thus, the average eGFR using all 4 markers (eGFRavg) could be more accurate than eGFRcr-cys in China.

STUDY DESIGN

Diagnostic test study.

SETTING & PARTICIPANTS

1,066 participants in Shanghai and Beijing with creatinine and cystatin C and 666 participants with all 4 filtration markers.

TESTS COMPARED

Index tests were previously developed equations for eGFR using creatinine, cystatin C, B2M, and BTP and combinations. The reference test was mGFR using plasma clearance of iohexol. We compared the performance of eGFRavg to eGFRcr-cys using the proportion of participants with errors in eGFR >30% of mGFR (1 - P30) and root mean square error (RMSE) of the regression of eGFR on mGFR on the logarithmic scale. We also compared classification and reclassification of mGFR categories using eGFRavg compared to eGFRcr-cys.

OUTCOMES

Accuracy was significantly better for eGFRavg (1 - P30 of 10.4% and RMSE of 0.214) compared to eGFRcr-cys (1 - P30 of 13.8% and RMSE of 0.232; P = 0.004 and P = 0.006, respectively). However, improvements in accuracy did not generally translate into significant improvement in classification or reclassification of mGFR categories.

LIMITATIONS

Study population may not be generalizable to clinical settings other than large urban medical centers in China.

CONCLUSIONS

A panel of endogenous filtration markers including B2M and BTP in addition to creatinine and cystatin C may improve GFR estimation in China. Further study is necessary to determine whether GFR estimation using B2M and BTP can be improved and whether these improvements lead to useful clinical applications.

Scatter to volume registration for model-free respiratory motion estimation from dynamic MRIs

Respiratory motion is one major complicating factor in many image acquisition applications and image-guided interventions. Existing respiratory motion estimation and compensation methods typically rely on breathing motion models learned from certain training data, and therefore may not be able to effectively handle intra-subject and/or inter-subject variations of respiratory motion. In this paper, we propose a respiratory motion compensation framework that directly recovers motion fields from sparsely spaced and efficiently acquired dynamic 2-D MRIs without using a learned respiratory motion model. We present a scatter-to-volume deformable registration algorithm to register dynamic 2-D MRIs with a static 3-D MRI to recover dense deformation fields. Practical considerations and approximations are provided to solve the scatter-to-volume registration problem efficiently. The performance of the proposed method was investigated on both synthetic and real MRI datasets, and the results showed significant improvements over the state-of-art respiratory motion modeling methods. We also demonstrated a potential application of the proposed method on MRI-based motion corrected PET imaging using hybrid PET/MRI.

Real-time 6DoF pose recovery from X-ray images using library-based DRR and hybrid optimization

PURPOSE

Real-time 6 degrees of freedom (6DoF) pose recovery and tracking from X-ray images is a key enabling technology for many interventional imaging applications. However, real-time 2D/3D registration is a very challenging problem because of the heavy computation in iterative digitally reconstructed radiograph (DRR) generation. In this paper, we propose a real-time 2D/3D registration framework using library-based DRRs to achieve high computational efficiency.

METHOD

The proposed method pre-computes a library of canonical DRRs and reconstructs library-based DRRs (libDRRs) during registration without online rendering. The transformation parameters are decoupled to 2 geometry-relevant and 4 geometry-irrelevant ones so that canonical DRRs only need to cover the variation of 2 geometry-relevant parameters, making it practical to be pre-computed and stored. The 2D/3D registration using libDRRs is then solved as a hybrid optimization problem, i.e., continuous in geometry-irrelevant parameters while discrete in geometry-relevant parameters.

RESULTS

On 5 fluoroscopic sequences with 246 frames acquired during animal studies with a transesophageal echocardiography (TEE) probe in the field of view, 6DoF tracking of the TEE probe using the proposed method achieved a mean target registration error in the projection direction (mTREproj) of 0.81 mm, a success rate of 100 % (defined as mTREproj [Formula: see text]2.5 mm), and a registration frame rate of 23.1 fps on a pure CPU-based implementation executed in a single thread.

CONCLUSION

Using libDRRs with a hybrid optimization can significantly improve the computational efficiency (up to tenfold) for 6DoF pose recovery and tracking with little degradation in robustness and accuracy, compared to conventional intensity-based 2D/3D registration using ray casting DRRs with a continuous optimization.

Extraordinary high ductility/strength of the interface designed bulk W-ZrC alloy plate at relatively low temperature

The refractory tungsten alloys with high ductility/strength/plasticity are highly desirable for a wide range of critical applications. Here we report an interface design strategy that achieves 8.5 mm thick W-0.5 wt. %ZrC alloy plates with a flexural strength of 2.5 GPa and a strain of 3% at room temperature (RT) and ductile-to-brittle transition temperature of about 100 °C. The tensile strength is about 991 MPa at RT and 582 MPa at 500 °C, as well as total elongation is about 1.1% at RT and as large as 41% at 500 °C, respectively. In addition, the W-ZrC alloy plate can sustain 3.3 MJ/m² thermal load without any cracks. This processing route offers the special coherent interfaces of grain/phase boundaries (GB/PBs) and the diminishing O impurity at GBs, which significantly strengthens GB/PBs and thereby enhances the ductility/strength/plasticity of W alloy. The design thought can be used in the future to prepare new alloys with higher ductility/strength.

INHIBITORY EFFECT OF TETRAMETHYLPYRAZINE ON HEPATOCELLULAR CARCINOMA: POSSIBLE ROLE OF APOPTOSIS AND CELL CYCLE ARREST

Hepatocellular carcinoma (HCC) is the fifth most common cancer. An important approach to control HCC is chemoprevention. This study aims at investigating the antitumor effect of Tetramethylpyrazine (TMP). Rats were injected with N-Nitrosodiethylamine (DEN) to establish HCC. Tumor development was observed. Liver function was evaluated. Apoptosis and cell cycle arrest-related makers and signaling cascades were determined by Western blot, RT-PCR and flow cytometric analysis. The administration of TMP could significantly inhibit tumor development in DEN-induced HCC rats, shown by reduced incidence of tumor, decreased number of tumor nodules and reduced maximal size of tumor. DEN-induced increase of aspartate aminotransferase, alanine aminotransferase, lactate dehydrogenase and alkaline phosphatase activities were significantly inhibited by TMP. TMP exhibited inhibitory effect on HCC through induction of apoptosis and cell cycle arrest in rats. TMP induced apoptosis through increasing Bax, decreasing Bcl-2, increasing the release of cytochrome c, and activating caspase, which consisted of the mitochondrial apoptotic pathway. TMP induced G2/M cell cycle arrest through down-regulation of cyclin B1/cdc2. In addition, inhibition of Akt and ERK signaling and the antioxidant activities of TMP may also contribute to its antitumor effect. These data provide new insight into the mechanisms underlying the antitumor effect of TMP.