Can low-frequency guided waves at the tibia paired with machine learning differentiate between healthy and osteopenic/osteoporotic subjects? A pilot study

Evaluation of fragility fracture risk using deep learning based on ultrasound radio frequency signal

BACKGROUND

It was essential to identify individuals at high risk of fragility fracture and prevented them due to the significant morbidity, mortality, and economic burden associated with fragility fracture. The quantitative ultrasound (QUS) showed promise in assessing bone structure characteristics and determining the risk of fragility fracture.

AIMS

To evaluate the performance of a multi-channel residual network (MResNet) based on ultrasonic radiofrequency (RF) signal to discriminate fragility fractures retrospectively in postmenopausal women, and compared it with the traditional parameter of QUS, speed of sound (SOS), and bone mineral density (BMD) acquired with dual X-ray absorptiometry (DXA).

METHODS

Using QUS, RF signal and SOS were acquired for 246 postmenopausal women. An MResNet was utilized, based on the RF signal, to categorize individuals with an elevated risk of fragility fracture. DXA was employed to obtain BMD at the lumbar, hip, and femoral neck. The fracture history of all adult subjects was gathered. Analyzing the odds ratios (OR) and the area under the receiver operator characteristic curves (AUC) was done to evaluate the effectiveness of various methods in discriminating fragility fracture.

RESULTS

Among the 246 postmenopausal women, 170 belonged to the non-fracture group, 50 to the vertebral group, and 26 to the non-vertebral fracture group. MResNet was competent to discriminate any fragility fracture (OR = 2.64; AUC = 0.74), Vertebral fracture (OR = 3.02; AUC = 0.77), and non-vertebral fracture (OR = 2.01; AUC = 0.69). After being modified by clinical covariates, the efficiency of MResNet was further improved to OR = 3.31-4.08, AUC = 0.81-0.83 among all fracture groups, which significantly surpassed QUS-SOS (OR = 1.32-1.36; AUC = 0.60) and DXA-BMD (OR = 1.23-2.94; AUC = 0.63-0.76).

CONCLUSIONS

This pilot cross-sectional study demonstrates that the MResNet model based on the ultrasonic RF signal shows promising performance in discriminating fragility fractures in postmenopausal women. When incorporating clinical covariates, the efficiency of the modified MResNet is further enhanced, surpassing the performance of QUS-SOS and DXA-BMD in terms of OR and AUC. These findings highlight the potential of the MResNet as a promising approach for fracture risk assessment. Future research should focus on larger and more diverse populations to validate these results and explore its clinical applications.

Bi-Directional Axial Transmission measurements applied in a clinical environment

Accurate measurement of cortical bone parameters may improve fracture risk assessment and help clinicians on the best treatment strategy. Patients at risk of fracture are currently detected using the current X-Ray gold standard DXA (Dual XRay Absorptiometry). Different alternatives, such as 3D X-Rays, Magnetic Resonance Imaging or Quantitative Ultrasound (QUS) devices, have been proposed, the latter having advantages of being portable and sensitive to mechanical and geometrical properties. The objective of this cross-sectional study was to evaluate the performance of a Bi-Directional Axial Transmission (BDAT) device used by trained operators in a clinical environment with older subjects. The device, positioned at one-third distal radius, provides two velocities: VFAS (first arriving signal) and VA0 (first anti-symmetrical guided mode). Moreover, two parameters are obtained from an inverse approach: Ct.Th (cortical thickness) and Ct.Po (cortical porosity), along with their ratio Ct.Po/Ct.Th. The areal bone mineral density (aBMD) was obtained using DXA at the femur and spine. One hundred and six patients (81 women, 25 men) from Marien Hospital and St. Anna Hospital (Herne, Germany) were included in this study. Age ranged from 41 to 95 years, while body mass index (BMI) ranged from 16 to 47 kg.m-2. Three groups were considered: 79 non-fractured patients (NF, 75±13years), 27 with non-traumatic fractures (F, 80±9years) including 14 patients with non-vertebral fractures (NVF, 84±7years). Weak to moderate significant Spearman correlations (R ranging from 0.23 to 0.53, p < 0.05) were found between ultrasound parameters and age, BMI. Using multivariate Partial Least Square discrimination analyses with Leave-One-Out Cross-Validation (PLS-LOOCV), we found the combination of VFAS and the ratio Ct.Po/Ct.Th to be predictive for all non traumatic fractures (F) with the odds ratio (OR) equals to 2.5 [1.6-3.4] and the area under the ROC curve (AUC) equal to 0.63 [0.62-0.65]. For the group NVF, combination of four parameters VA0. Ct.Th, Ct.Po and Ct.Po/Ct.Po, along with age provides a discrimination model with OR and AUC equals to 7.5 [6.0-9.1] and 0.75 [0.73-0.76]. When restricted to a smaller population (87 patients) common to both BDAT and DXA, BDAT ORs and AUCs are comparable or slightly higher to values obtained with DXA. The fracture risk assessment by BDAT method in older patients, in a clinical setting, suggests the benefit of the affordable and transportable device for the routine use.

Ultrasonic Guided Waves in Bone: A Decade of Advancement in Review

The use of guided wave ultrasonography as a means to assess cortical bone quality has been a significant practice in bone quantitative ultrasound for more than 20 years. In this article, the key developments within the technology of ultrasonic guided waves (UGW) in long bones during the past decade are documented. The covered topics include data acquisition configurations available for measuring bone guided waveforms, signal processing techniques applied to bone UGW, numerical modeling of ultrasonic wave propagation in cortical long bones, formulation of inverse approaches to extract bone properties from observed ultrasonic signals, and clinical studies to establish the technology's application and efficacy. The review concludes by highlighting specific challenging problems and future research directions. In general, the primary purpose of this work is to provide a comprehensive overview of bone guided-wave ultrasound, especially for newcomers to this scientific field.

Current Applications of Machine Learning in Spine: From Clinical View

STUDY DESIGN

Narrative review.

OBJECTIVES

This review aims to present current applications of machine learning (ML) in spine domain to clinicians.

METHODS

We conducted a comprehensive PubMed search of peer-reviewed articles that were published between 2006 and 2020 using terms (spine, spinal, lumbar, cervical, thoracic, machine learning) to examine ML in spine. Then exclude research of other domain, case report, review or meta-analysis, and which without available abstract or full text.

RESULTS

Total 1738 articles were retrieved from database, and 292 studies were finally included. Key findings of current applications were compiled and summarized in this review. Main clinical applications of those techniques including image processing, diagnosis, decision supporting, operative assistance, rehabilitation, surgery outcomes, complications, hospitalization and cost.

CONCLUSIONS

ML had achieved excellent performance and hold immense potential in spine. ML could help clinical staff to improve medical level, enhance work efficiency, and reduce adverse events. However more randomized controlled trials and improvement of interpretability are essential to clinicians accepting models' assistance in real work.

Can machine learning predict pharmacotherapy outcomes? An application study in osteoporosis

BACKGROUND AND OBJECTIVE

The specific aim of this study is to develop machine learning models as a clinical approach for personalized treatment of osteoporosis. The model performance on outcome prediction was compared between four machine learning algorithms.

METHODS

Retrospective, electronic clinical data for patients with suspected or confirmed osteoporosis treated at Wan Fang Hospital between 2011 to 2018 were used as inputs for building the following predictive machine learning models,i.e., artificial neural network (ANN), random forest (RF), support vector machine (SVM) and logistic regression (LR) models. The predicted outcome was defined as an increase/decrease in T-score after treatment. A genetic algorithm was employed to select relevant variables as input features for each model; the leave-one-out method was applied for model building and internal validation. The model with best performance was selected by a separate set of testing. Area under the receiver operating characteristic curve, accuracy, precision, sensitivity and F1 score were calculated to evaluate model performance. Main analysis for all the patients with subclinical or confirmed osteoporosis and subgroup analysis for the patients with confirmed osteoporosis (T score < -2.5) were carried out in this study.

RESULTS

A genetic algorithm was employed to select 12 to 18 features from all 33 variables for the four models. No difference was found in accuracy (ANN, 71.7%; LR, 70.0%; RF, 75.0%; SVM, 66.7%), precision (ANN, 80.0%; LR, 59.3%; RF, 70.0%; SVM, 63.6%), and AUC (ANN, 0.709; LR, 0.731; RF, 0.719; SVM, 0.702) among the ANN, LR, RF and SVM models. Main analysis in performance revealed significant recall in the LR model, as compared to ANN and SVM model; while subgroup revealed significant recall in ANN model, compared to LR and SVM model.

CONCLUSIONS

Machine learning-based models hold potential in forecasting the outcomes of treatment for osteoporosis via early initiation of first-line therapy for patients with subclinical disease; or a switch to second-line treatment for patients with a high risk of impending treatment failure. This convenient approach can assist clinicians in adjusting treatment tailored to individual patient for prevention of disease progression or ineffective therapy.

Bone Mineral Content Estimation in People Living with HIV: Prediction and Validation of Sex-Specific Anthropometric Models

People living with HIV (PWH) experience an accelerated reduction in bone mineral content (BMC), and a high risk of osteopenia and osteoporosis. Anthropometry is an accurate and low-cost method that can be used to monitor changes in body composition in PWH. To date, no studies have used anthropometry to estimate BMC in PWH. To propose and validate sex-specific anthropometric models to predict BMC in PWH. This cross-sectional study enrolled 104 PWH (64 males) aged >18 years at a local university hospital. BMC was measured using dual energy X-ray absorptiometry (DXA). Anthropometric measures were collected. We used linear regression analysis to generate the models. Cross-validations were conducted using the "leave one out", from the predicted residual error sum of squares (PRESS) method. Bland-Altman plots were used to explore distributions of errors. We proposed models with high coefficient of determination and reduced standard error of estimate for males (r² = 0.70; SEE = 199.97 g; Q²_PRESS = 0.67; SEE_PRESS = 208.65 g) and females (r² = 0.65; SEE = 220.96 g; Q²_PRESS = 0.62; SEE_PRESS = 221.90 g). Our anthropometric predictive models for BMC are valid, practical, and a low-cost alternative to monitoring bone health in PWH.

Design and calibration of a Tonpilz transducer for low frequency medical ultrasound tomography

The design and performance of a transducer for low frequency ultrasound tomography is presented, motivated by recent research demonstrating that acoustic waves transmitting at frequencies between 10 kHz and 750 kHz penetrate the lungs and may be useful for thoracic imaging. An adaptation of the traditional Tonpilz design was developed, vibrational amplitude and electrical impedance were measured, and an optimal frequency was determined. The design is found to meet the desired mechanical, electrical, and safety specifications. Thus, it was considered a promising option for the target application of pulmonary imaging with ultrasound computed tomography between 50 and 200 kHz; highest efficiency achieved around 125 kHz and 156 kHz, and beam divergence of 40°.

Meta-Learning Analysis of Ultrasonic Guided Waves for Coated Cortical Bone Characterization

Due to its sensitivity to geometrical and mechanical properties of waveguides, ultrasonic guided waves (UGWs) propagating in cortical bones play an important role in the early diagnosis of osteoporosis. However, as impacts of overlaid soft tissues are complex, it remains challenging to retrieve bone properties accurately. Meta-learning, i.e., learning to learn, is capable of extracting transferable features from a few data and, thus, suitable to capture potential characteristics, leading to accurate bone assessment. In this study, we investigate the feasibility to apply the multichannel identification neural network (MCINN) to estimate the thickness and bulk velocities of coated cortical bone. It minimizes the effects of soft tissue by extracting specific features of UGW, which shares the same cortical properties, while the overlaid soft tissue varies. Distinguished from most reported methods, this work moves from the hand-design inversion scheme to data-driven assessment by automatically mapping features of UGW to the space of bone properties. The MCINN was trained and validated using simulated datasets produced by the finite-difference time-domain (FDTD) method and then applied to experimental data obtained from cortical bovine bone plates overlaid with soft tissue mimics. A good match was found between experimental trajectories and theoretical dispersion curves. The results demonstrated that the proposed method was feasible to assess the thickness of coated cortical bone plates.

A hierarchical opportunistic screening model for osteoporosis using machine learning applied to clinical data and CT images

Background

Osteoporosis is a common metabolic skeletal disease and usually lacks obvious symptoms. Many individuals are not diagnosed until osteoporotic fractures occur. Bone mineral density (BMD) measured by dual-energy X-ray absorptiometry (DXA) is the gold standard for osteoporosis detection. However, only a limited percentage of people with osteoporosis risks undergo the DXA test. As a result, it is vital to develop methods to identify individuals at-risk based on methods other than DXA.

Results

We proposed a hierarchical model with three layers to detect osteoporosis using clinical data (including demographic characteristics and routine laboratory tests data) and CT images covering lumbar vertebral bodies rather than DXA data via machine learning. 2210 individuals over age 40 were collected retrospectively, among which 246 individuals’ clinical data and CT images are both available. Irrelevant and redundant features were removed via statistical analysis. Consequently, 28 features, including 16 clinical data and 12 texture features demonstrated statistically significant differences (p < 0.05) between osteoporosis and normal groups. Six machine learning algorithms including logistic regression (LR), support vector machine with radial-basis function kernel, artificial neural network, random forests, eXtreme Gradient Boosting and Stacking that combined the above five classifiers were employed as classifiers to assess the performances of the model. Furthermore, to diminish the influence of data partitioning, the dataset was randomly split into training and test set with stratified sampling repeated five times. The results demonstrated that the hierarchical model based on LR showed better performances with an area under the receiver operating characteristic curve of 0.818, 0.838, and 0.962 for three layers, respectively in distinguishing individuals with osteoporosis and normal BMD.

Conclusions

The proposed model showed great potential in opportunistic screening for osteoporosis without additional expense. It is hoped that this model could serve to detect osteoporosis as early as possible and thereby prevent serious complications of osteoporosis, such as osteoporosis fractures.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12859-022-04596-z.

Fully automated radiomic screening pipeline for osteoporosis and abnormal bone density with a deep learning-based segmentation using a short lumbar mDixon sequence

BACKGROUND

Although lumbar bone marrow fat fraction (BMFF) has been demonstrated to be predictive of osteoporosis, its utility is limited by the requirement of manual segmentation. Additionally, quantitative features beyond simple BMFF average remain to be explored. In this study, we developed a fully automated radiomic pipeline using deep learning-based segmentation to detect osteoporosis and abnormal bone density (ABD) using a <20 s modified Dixon (mDixon) sequence.

METHODS

In total, 222 subjects underwent quantitative computed tomography (QCT) and lower back magnetic resonance imaging (MRI). Bone mineral density (BMD) were extracted from L1-L3 using QCT as the reference standard; 206 subjects (48.8±14.9 years old, 140 females) were included in the final analysis, and were divided temporally into the training/validation set (142/64 subjects). A deep-learning network was developed to perform automated segmentation. Radiomic models were built using the same training set to predict ABD and osteoporosis using the mDixon maps. The performance was evaluated using the temporal validation set comprised of 64 subjects, along with the automated segmentation. Additional 25 subjects (56.1±8.8 years, 14 females) from another site and a different scanner vendor was included as independent validation to evaluate the performance of the pipeline.

RESULTS

The automated segmentation achieved an outstanding mean dice coefficient of 0.912±0.062 compared to manual in the temporal validation. Task-based evaluation was performed in the temporal validation set, for predicting ABD and osteoporosis, the area under the curve, sensitivity, specificity, and accuracy were 0.925/0.899, 0.923/0.667, 0.789/0.873, 0.844/0.844, respectively. These values were comparable to that of manual segmentation. External validation (cross-vendor) was also performed; the area under the curve, sensitivity, specificity, and accuracy were 0.688/0.913, 0.786/0.857, 0.545/0.944, 0.680/0.920 for ABD and osteoporosis prediction, respectively.

CONCLUSIONS

Our work is the first attempt using radiomics to predict osteoporosis with BMFF map, and the deep-learning based segmentation will further facilitate the clinical utility of the pipeline as a screening tool for early detection of ABD.

Fourier-Domain Ultrasonic Imaging of Cortical Bone Based on Velocity Distribution Inversion

There is a significant acoustic impedance contrast between the cortical bone and the surrounding soft tissue, resulting in difficulty for ultrasound penetration into bone tissue with high frequency. It is challenging for the conventional pulse-echo modalities to give accurate cortical bone images using uniform sound velocity model. To overcome these limitations, an ultrasound imaging method called full-matrix Fourier-domain synthetic aperture based on velocity inversion (FM-FDSA-VI) was developed to provide accurate cortical bone images. The dual linear arrays were located on the upper and lower sides of the imaging region. After full-matrix acquisition with two identical linear array probes facing each other, travel-time inversion was used to estimate the velocity distribution in advance. Then, full-matrix Fourier-domain synthetic aperture (FM-FDSA) imaging based on the estimated velocity model was applied twice to image the cortical bone, utilizing the data acquired from top and bottom linear array, respectively. Finally, to further improve the image quality, the two images were merged to give the ultimate result. The performance of the method was verified by two simulated models and two bone phantoms (i.e., regular and irregular hollow bone phantom). The mean relative errors of estimated sound velocity in the region-of-interest (ROI) are all below 12%, and the mean errors of cortical section thickness are all less than 0.3 mm. Compared to the conventional synthetic aperture (SA) imaging, the FM-FDSA-VI method is able to accurately image cortical bone with respect to the structure. Moreover, the result of irregular bone phantom was close to the image scanned by microcomputed tomography ( μ CT) in terms of macro geometry and thickness. It is demonstrated that the proposed FM-FDSA-VI method is an efficient way for cortical bone ultrasonic imaging.

Machine Learning Solutions for Osteoporosis-A Review

Osteoporosis and its clinical consequence, bone fracture, is a multifactorial disease that has been the object of extensive research. Recent advances in machine learning (ML) have enabled the field of artificial intelligence (AI) to make impressive breakthroughs in complex data environments where human capacity to identify high-dimensional relationships is limited. The field of osteoporosis is one such domain, notwithstanding technical and clinical concerns regarding the application of ML methods. This qualitative review is intended to outline some of these concerns and to inform stakeholders interested in applying AI for improved management of osteoporosis. A systemic search in PubMed and Web of Science resulted in 89 studies for inclusion in the review. These covered one or more of four main areas in osteoporosis management: bone properties assessment (n = 13), osteoporosis classification (n = 34), fracture detection (n = 32), and risk prediction (n = 14). Reporting and methodological quality was determined by means of a 12-point checklist. In general, the studies were of moderate quality with a wide range (mode score 6, range 2 to 11). Major limitations were identified in a significant number of studies. Incomplete reporting, especially over model selection, inadequate splitting of data, and the low proportion of studies with external validation were among the most frequent problems. However, the use of images for opportunistic osteoporosis diagnosis or fracture detection emerged as a promising approach and one of the main contributions that ML could bring to the osteoporosis field. Efforts to develop ML-based models for identifying novel fracture risk factors and improving fracture prediction are additional promising lines of research. Some studies also offered insights into the potential for model-based decision-making. Finally, to avoid some of the common pitfalls, the use of standardized checklists in developing and sharing the results of ML models should be encouraged. © 2021 American Society for Bone and Mineral Research (ASBMR).

Deep Learning Analysis of Ultrasonic Guided Waves for Cortical Bone Characterization

Ultrasonic guided waves (UGWs) propagating in the long cortical bone can be measured via the axial transmission method. The characterization of long cortical bone using UGW is a multiparameter inverse problem. The optimal solution of the inverse problem often includes a complex solving process. Deep neural networks (DNNs) are essentially powerful multiparameter predictors based on universal approximation theorem, which are suitable for solving parameter predictions in the inverse problem by constructing the mapping relationship between UGW and cortical bone material parameters. In this study, we investigate the feasibility of applying the multichannel crossed convolutional neural network (MCC-CNN) to simultaneously estimate cortical thickness and bulk velocities (longitudinal and transverse). Unlike the multiparameter estimation in most previous studies, the technique mentioned in this work avoids solving a multiparameter optimization problem directly. The finite-difference time-domain (FDTD) method is performed to obtain the simulated UGW array signals for training the MCC-CNN. The network that is exclusively trained on simulated data sets can predict cortical parameters from the experimental UGW data. The proposed method is confirmed by using FDTD simulation signals and experimental data obtained from four bone-mimicking plates and from ten ex vivo bovine cortical bones. The estimated root-mean-squared error (RMSE) in the simulated test data for the longitudinal bulk velocity ( V_L ), transverse bulk velocity ( V_T ), and cortical thickness (Th) is 97 m/s, 53 m/s, and 0.089 mm, respectively. The predicted RMSE in the bone-mimicking phantom experiments for V_L|| , V_T|| , and Th is 120 m/s, 80 m/s, and 0.14 mm, respectively. The experimental dispersion trajectories are matched with the theoretical dispersion curves calculated by the predicted parameters in ex vivo bovine cortical bone experiments. Our proposed method demonstrates a feasible approach for the accurate evaluation of long cortical bones based on UGW.

Artificial intelligence, osteoporosis and fragility fractures

PURPOSE OF REVIEW

Artificial intelligence tools have found new applications in medical diagnosis. These tools have the potential to capture underlying trends and patterns, otherwise impossible with previous modeling capabilities. Machine learning and deep learning models have found a role in osteoporosis, both to model the risk of fragility fracture, and to help with the identification and segmentation of images.

RECENT FINDINGS

Here we survey the latest research in the artificial intelligence application to the prediction of osteoporosis that has been published between January 2017 and March 2019. Around half of the articles that are covered here predict (by classification or regression) an indicator of osteoporosis, such as bone mass or fragility fractures; the other half of studies use tools for automatic segmentation of the images of patients with or at risk of osteoporosis. The data for these studies include diverse signal sources: acoustics, MRI, CT, and of course, X-rays.

SUMMARY

New methods for automatic image segmentation, and prediction of fracture risk show promising clinical value. Though these recent developments have had a successful initial application to osteoporosis research, their development is still under improvement, such as accounting for positive/negative class bias. We urge care when reporting accuracy metrics, and when comparing such metrics between different studies.

Artificial neural network to estimate micro-architectural properties of cortical bone using ultrasonic attenuation: A 2-D numerical study

The goal of this study is to estimate micro-architectural parameters of cortical porosity such as pore diameter (φ), pore density (ρ) and porosity (ν) of cortical bone from ultrasound frequency dependent attenuation using an artificial neural network (ANN). First, heterogeneous structures with controlled pore diameters and pore densities (mono-disperse) were generated, to mimic simplified structure of cortical bone. Then, more realistic structures were obtained from high resolution CT scans of human cortical bone. 2-D finite-difference time-domain simulations were conducted to calculate the frequency-dependent attenuation in the 1-8 MHz range. An ANN was then trained with the ultrasonic attenuation at different frequencies as the input feature vectors while the output was set as the micro-architectural parameters (pore diameter, pore density and porosity). The ANN is composed of three fully connected dense layers with 24, 12 and 6 neurons, connected to the output layer. The dataset was trained over 6000 epochs with a batch size of 16. The trained ANN exhibits the ability to predict the micro-architectural parameters with high accuracy and low losses. ANN approaches could potentially be used as a tool to help inform physics-based modelling of ultrasound propagation in complex media such as cortical bone. This will lead to the solution of inverse-problems to retrieve bone micro-architectural parameters from ultrasound measurements for the non-invasive diagnosis and monitoring osteoporosis.

Classification of Micro-Damage in Piezoelectric Ceramics Using Machine Learning of Ultrasound Signals

Ultrasound based structural health monitoring of piezoelectric material is challenging if a damage changes at a microscale over time. Classifying geometrically similar damages with a difference in diameter as small as 100 μm is difficult using conventional sensing and signal analysis approaches. Here, we use an unconventional ultrasound sensing approach that collects information of the entire bulk of the material and investigate the applicability of machine learning approaches for classifying such similar defects. Our results show that appropriate feature design combined with simple k-nearest neighbor classifier can provide up to 98% classification accuracy even though conventional features for time-series data and a variety of classifiers cannot achieve close to 70% accuracy. The newly proposed hybrid feature, which combines frequency domain information in the form of power spectral density and time domain information in the form of sign of slope change, is a suitable feature for achieving the best classification accuracy on this challenging problem.