Interpretable Deep-Learning Approaches for Osteoporosis Risk Screening and Individualized Feature Analysis Using Large Population-Based Data: Model Development and Performance Evaluation

Development and validation of explainable machine learning models for female hip osteoporosis using electronic health records

BACKGROUND

Hip fractures are associated with reduced mobility, and higher morbidity, mortality, and healthcare costs. Approximately 90% of hip fractures in the elderly are associated with osteoporosis, making it particularly important to screen the population for hip osteoporosis and intervene early. Dual-energy X-ray absorptiometry (DXA) has limited accessibility, so predictive models for hip osteoporosis that do not use bone mineral density (BMD) data are essential. We aimed to develop and validate prediction models for female hip osteoporosis using electronic health records without BMD data.

METHODS

This retrospective study used anonymized medical electronic records, from September 2013 to November 2023, from the Health Management Center of the Second Xiangya Hospital. A total of 8039 women were included in the derivation dataset. The set was then randomized into a 75% training dataset and a 25% testing dataset. Four algorithms for feature selection were used to identify predictors of osteoporosis. The identified predictors were then used to train and optimize eight machine learning models. The models were tuned using 5-fold cross-validation to assess model performance in the testing dataset and the independent validation dataset from the National Health and Nutrition Examination Surveys (NHANES). The SHapley Additive explanation (SHAP) method was used to rank feature importance and explain the final model.

RESULTS

A combination of the Boruta, LASSO, varSelRF, and RFE methods identified systolic blood pressure, red blood cell count, glycohemoglobin, alanine aminotransferase, aspartate aminotransferase, uric acid, age, and body mass index as the most important predictors of osteoporosis in women. The XGBoost model outperformed the other models, with an Area Under the Curve (AUC) of 0.805 (95%CI: 0.779-0.831), and a moderate sensitivity of 0.706. The externally validated XGBoost model had an AUC of 0.811 (95% CI: 0.793-0.828), with a moderate sensitivity of 0.775.

CONCLUSIONS

The XGBoost model demonstrates high identification performance even without questionnaire data, out-performing both the traditional the logistic regression model and the OSTA model. It can be integrated into routine clinical workflows to identify females at high risk for osteoporosis.

Exploring the classification and treatment of osteoporosis from the perspectives of natural medicines, molecular targets, and symptom clusters

Osteoporosis (OP) is a metabolic bone disease characterized by reduced bone density and fragility, impairing quality of life. Traditional treatments often overlook symptoms like back and joint pain, increasing burden. This study aims to map relationships between natural medicines, targets, and symptom clusters, demonstrating their effectiveness in personalized OP treatment to enhance clinical strategies and self-assessment. We used compounds and targets, applying Summary data-based Mendelian Randomisation (SMR) analysis for biological process and molecular function enrichment. Additionally, we employed Phenome-Wide Association Studies (PheWAS) to select two natural drugs-Rhizoma Drynariae (RD) and Lycii Fructus (LF)-for case analysis. The study found that RD primarily improves symptoms such as indigestion, constipation, fatigue, polyuria, and depression, while LF significantly ameliorates symptoms related to the nervous and muscular systems, such as hoarseness, dizziness, vertigo, and fever symptoms. This analysis successfully differentiated two groups of symptoms and precisely constructed a logical chain among "natural Medicines-molecular tArGets-Illness-symptom Clusters" (MAGIC chain) achieving a refined classification of OP. The results of this study support the effectiveness of implementing personalized medical strategies in the treatment of OP, providing a scientific basis for the clinical application of natural medicines and patient self-management.

PrOsteoporosis: predicting osteoporosis risk using NHANES data and machine learning approach

OBJECTIVES

Osteoporosis, prevalent among the elderly population, is primarily diagnosed through bone mineral density (BMD) testing, which has limitations in early detection. This study aims to develop and validate a machine learning approach for osteoporosis identification by integrating demographic data, laboratory and questionnaire data, offering a more practical and effective screening alternative.

METHODS

In this study, data from the National Health and Nutrition Examination Survey were analyzed to explore factors linked to osteoporosis. After cleaning, 8766 participants with 223 variables were studied. Minimum Redundancy Maximum Relevance and SelectKBest were employed to select the import features. Four Machine learning algorithms (RF, NN, LightGBM and XGBoost.) were applied to examine osteoporosis, with performance comparisons made. Data balancing was done using SMOTE, and metrics like F1 score, and AUC were evaluated for each algorithm.

RESULTS

The LightGBM model outperformed others with an F1 score of 0.914, an MCC of 0.831, and an AUC of 0.970 on the training set. On the test set, it achieved an F1 score of 0.912, an MCC of 0.826, and an AUC of 0.972. Top predictors for osteoporosis were height, age, and sex.

CONCLUSIONS

This study demonstrates the potential of machine learning models in assessing an individual's risk of developing osteoporosis, a condition that significantly impacts quality of life and imposes substantial healthcare costs. The superior performance of the LightGBM model suggests a promising tool for early detection and personalized prevention strategies. Importantly, identifying height, age, and sex as top predictors offers critical insights into the demographic and physiological factors that clinicians should consider when evaluating patients' risk profiles.

A simple and user-friendly machine learning model to detect osteoporosis in health examination populations in Southern Taiwan

Background

Osteoporosis is a growing public health concern in aging populations such as Taiwan, where limited utilization of dual-energy X-ray absorptiometry (DXA) often leads to underdiagnosis and even delayed treatment. Therefore, we leveraged machine learning (ML) and aimed to develop a simple and easily accessible model that effectively identifies individuals at high risk of osteoporosis.

Methods

This retrospective analysis enrolled 5510 men aged ≥50 years and 4720 postmenopausal women who underwent DXA at the Kaohsiung Veterans General Hospital, with another cohort of 610 men and 523 women for validation. We developed separate models for men and women using decision trees, random forests, support vector machines, k-nearest neighbors, extreme gradient boosting, and artificial neural networks (ANNs) to predict osteoporosis. Furthermore, we compared each model with the traditional Osteoporosis Self-Assessment Tool for Asians (OSTA) model.

Results

We identified age, height, weight, and BMI as variables for our prediction model and evaluated the model's performance using the area under the receiver operating characteristic curve (AUC). The ANN model significantly outperformed the OSTA model and all the other ML models for both men and women (AUC: 0.67 for men; 0.77 for women). The validation data for the ANN model showed similar AUCs for both men and women.

Conclusion

This study developed ML models to help identify individuals at high risk of osteoporosis in postmenopausal women and men aged ≥50 years in southern Taiwan. Our ML models, especially the ANN model, surpassed the OSTA model and consistently performed well across different populations.

Efficacy and Mechanism of Highly Active Umbilical Cord Mesenchymal Stem Cells in the Treatment of Osteoporosis in Rats

BACKGROUND

Osteoporosis increases bone brittleness and the risk of fracture. Umbilical cord mesenchymal stem cell (UCMSC) treatment is effective, but how to improve the biological activity and clinical efficacy of UCMSCs has not been determined.

METHODS

A rat model of osteoporosis was induced with dexamethasone sodium phosphate. Highly active umbilical cord mesenchymal stem cells (HA-UCMSCs) and UCMSCs were isolated, cultured, identified, and infused intravenously once at a dose of 2.29 × 106 cells/kg. In the 4th week of treatment, bone mineral density (BMD) was evaluated via cross-micro-CT, tibial structure was observed via HE staining, osteogenic differentiation of bone marrow mesenchymal stem cells (BMMSCs) was examined via alizarin red staining, and carboxy-terminal cross-linked telopeptide (CTX), nuclear factor-κβ ligand (RANKL), procollagen type 1 N-terminal propeptide (PINP) and osteoprotegerin (OPG) levels were investigated via enzyme-linked immunosorbent assays (ELISAs). BMMSCs were treated with 10-6 mol/L dexamethasone and cocultured with HA-UCMSCs and UCMSCs in transwells. The osteogenic and adipogenic differentiation of BMMSCs was subsequently examined through directional induction culture. The protein expression levels of WNT, β-catenin, RUNX2, IFN-γ and IL-17 in the bone tissue were measured via Western blotting.

RESULTS

The BMD in the healthy group was higher than that in the model group. Both UCMSCs and HA-UCMSCs exhibited a fusiform morphology; swirling growth; high expression of CD73, CD90 and CD105; and low expression of CD34 and CD45 and could differentiate into adipocytes, osteoblasts and chondrocytes, while HA-UCMSCs were smaller in size; had a higher nuclear percentage; and higher differentiation efficiency. Compared with those in the model group, the BMD increased, the bone structure improved, the trabecular area, number, and perimeter increased, the osteogenic differentiation of BMMSCs increased, RANKL expression decreased, and PINP expression increased after UCMSC and HA-UCMSC treatment for 4 weeks. Furthermore, the BMD, trabecular area, number and perimeter, calcareous nodule counts, and OPG/RANKL ratio were higher in the HA-UCMSC treatment group than in the UCMSC treatment group. The osteogenic and adipogenic differentiation of dexamethasone-treated BMMSCs was enhanced after the coculture of UCMSCs and HA-UCMSCs, and the HA-UCMSC group exhibited better effects than the UCMSC coculture group. The protein expression of WNT, β-catenin, and runx2 was upregulated, and IFN-γ and IL-17 expression was downregulated after UCMSC and HA-UCMSC treatment.

CONCLUSION

HA-UCMSCs have a stronger therapeutic effect on osteoporosis compared with that of UCMSCs. These effects include an improved bone structure, increased BMD, an increased number and perimeter of trabeculae, and enhanced osteogenic differentiation of BMMSCs via activation of the WNT/β-catenin pathway and inhibition of inflammation.

ELW-CNN: An extremely lightweight convolutional neural network for enhancing interoperability in colon and lung cancer identification using explainable AI

Cancer is a condition in which cells in the body grow uncontrollably, often forming tumours and potentially spreading to various areas of the body. Cancer is a hazardous medical case in medical history analysis. Every year, many people die of cancer at an early stage. Therefore, it is necessary to accurately and early identify cancer to effectively treat and save human lives. However, various machine and deep learning models are effective for cancer identification. Therefore, the effectiveness of these efforts is limited by the small dataset size, poor data quality, interclass changes between lung squamous cell carcinoma and adenocarcinoma, difficulties with mobile device deployment, and lack of image and individual-level accuracy tests. To overcome these difficulties, this study proposed an extremely lightweight model using a convolutional neural network that achieved 98.16% accuracy for a large lung and colon dataset and individually achieved 99.02% for lung cancer and 99.40% for colon cancer. The proposed lightweight model used only 70 thousand parameters, which is highly effective for real-time solutions. Explainability methods such as Grad-CAM and symmetric explanation highlight specific regions of input data that affect the decision of the proposed model, helping to identify potential challenges. The proposed models will aid medical professionals in developing an automated and accurate approach for detecting various types of colon and lung cancer.

Explainable Deep Learning Approaches for Risk Screening of Periodontitis

Several pieces of evidence have been reported regarding the association between periodontitis and systemic diseases. Despite the emphasized significance of prevention and early diagnosis of periodontitis, there is still a lack of a clinical tool for early screening of this condition. Therefore, this study aims to use explainable artificial intelligence (XAI) technology to facilitate early screening of periodontitis. This is achieved by analyzing various clinical features and providing individualized risk assessment using XAI. We used 1,012 variables for a total of 30,465 participants data from National Health and Nutrition Examination Survey (NHANES). After preprocessing, 9,632 and 5,601 participants were left for all age groups and the over 50 y age group, respectively. They were used to train deep learning and machine learning models optimized for opportunistic screening and diagnosis analysis of periodontitis based on Centers for Disease Control and Prevention/ American Academy of Pediatrics case definition. Local interpretable model-agnostic explanations (LIME) were applied to evaluate potential associated factors, including demographic, lifestyle, medical, and biochemical factors. The deep learning models showed area under the curve values of 0.858 ± 0.011 for the opportunistic screening and 0.865 ± 0.008 for the diagnostic dataset, outperforming baselines. By using LIME, we elicited important features and assessed the combined impact and interpretation of each feature on individual risk. Associated factors such as age, sex, diabetes status, tissue transglutaminase, and smoking status have emerged as crucial features that are about twice as important than other features, while arthritis, sleep disorders, high blood pressure, cholesterol levels, and overweight have also been identified as contributing factors to periodontitis. The feature contribution rankings generated with XAI offered insights that align well with clinically recognized associated factors for periodontitis. These results highlight the utility of XAI in deep learning-based associated factor analysis for detecting clinically associated factors and the assistance of XAI in developing early detection and prevention strategies for periodontitis in medical checkups.

Development and reporting of artificial intelligence in osteoporosis management

An abundance of medical data and enhanced computational power have led to a surge in artificial intelligence (AI) applications. Published studies involving AI in bone and osteoporosis research have increased exponentially, raising the need for transparent model development and reporting strategies. This review offers a comprehensive overview and systematic quality assessment of AI articles in osteoporosis while highlighting recent advancements. A systematic search in the PubMed database, from December 17, 2020 to February 1, 2023 was conducted to identify AI articles that relate to osteoporosis. The quality assessment of the studies relied on the systematic evaluation of 12 quality items derived from the minimum information about clinical artificial intelligence modeling checklist. The systematic search yielded 97 articles that fell into 5 areas; bone properties assessment (11 articles), osteoporosis classification (26 articles), fracture detection/classification (25 articles), risk prediction (24 articles), and bone segmentation (11 articles). The average quality score for each study area was 8.9 (range: 7-11) for bone properties assessment, 7.8 (range: 5-11) for osteoporosis classification, 8.4 (range: 7-11) for fracture detection, 7.6 (range: 4-11) for risk prediction, and 9.0 (range: 6-11) for bone segmentation. A sixth area, AI-driven clinical decision support, identified the studies from the 5 preceding areas that aimed to improve clinician efficiency, diagnostic accuracy, and patient outcomes through AI-driven models and opportunistic screening by automating or assisting with specific clinical tasks in complex scenarios. The current work highlights disparities in study quality and a lack of standardized reporting practices. Despite these limitations, a wide range of models and examination strategies have shown promising outcomes to aid in the earlier diagnosis and improve clinical decision-making. Through careful consideration of sources of bias in model performance assessment, the field can build confidence in AI-based approaches, ultimately leading to improved clinical workflows and patient outcomes.

A few-shot learning framework for the diagnosis of osteopenia and osteoporosis using knee X-ray images

OBJECTIVE

We developed a few-shot learning (FSL) framework for the diagnosis of osteopenia and osteoporosis in knee X-ray images.

METHODS

Computer vision models containing deep convolutional neural networks were fine-tuned to enable generalization from natural images (ImageNet) to chest X-ray images (normal vs. pneumonia, base images). Then, a series of automated machine learning classifiers based on the Euclidean distances of base images were developed to make predictions for novel images (normal vs. osteopenia vs. osteoporosis). The performance of the FSL framework was compared with that of junior and senior radiologists. In addition, the gradient-weighted class activation mapping algorithm was used for visual interpretation.

RESULTS

In Cohort #1, the mean accuracy (0.728) and sensitivity (0.774) of the FSL models were higher than those of the radiologists (0.512 and 0.448). A diagnostic pipeline of FSL model (first)-radiologists (second) achieved better performance (0.653 accuracy, 0.582 sensitivity, and 0.816 specificity) than radiologists alone. In Cohort #2, the diagnostic pipeline also showed improved performance.

CONCLUSIONS

The FSL framework yielded practical performance with respect to the diagnosis of osteopenia and osteoporosis in comparison with radiologists. This retrospective study supports the use of promising FSL methods in computer-aided diagnosis tasks involving limited samples.

Proximal humeral bone density assessment and prediction analysis using machine learning techniques: An innovative approach in medical research

Background

Patients with fractures of the proximal humerus often local complications and failures attributed to osteoporosis. Currently, there is a lack of straight forward screening methods for assessing the extent of local osteoporosis in the proximal humerus. This study utilizes machine learning techniques to establish a diagnostic approach for evaluating local osteoporosis by analyzing the patient's demographic data, bone density, and X-ray ratio of the proximal humerus.

Methods

A cohort comprising a total of 102 hospitalized patients admitted during the period spanning from 2021 to 2023 underwent random selection procedures. Resulting in exclusion of 5 patients while enrolling 97 patients for analysis encompassing patient demographics, shoulder joint anteroposterior radiographs, and bone density information. Using the modified Tingart index methodology involving multiple measurements denoted as M1 through M4 obtained from humeral shafts. Within this cohort comprised 76 females (78.4 %) and 21 males (21.6 %), with an average age of 73.0 years (range: 43-98 years). There were 25 cases with normal bone density, 35 with osteopenia, and 37 with osteoporosis. Machine learning techniques were used to randomly divide the 97 cases into training (n = 59) and validation (n = 38) sets with a ratio of 6:4 using stratified random sampling. A decision tree model was built in the training set, and significant diagnostic indicators were selected, with the performance of the decision tree evaluated using the validation set. Multinomial logistic regression methods were used to verify the strength of the relationship between the selected indicators and osteoporosis.

Results

The decision tree identified significant diagnostic indicators as the humeral shaft medullary cavity ratio M2/M4, age, and gender. M2/M4 ≥ 1.13 can be used as an important screening criterion; M2/M4 < 1.13 was predicted as local osteoporosis; M2/M4 ≥ 1.13 and age ≥83 years were also predicted as osteoporosis. M2/M4 ≥ 1.13 and age <64 years or males aged between 64 and 83 years were predicted as the normal population; M2/M4 ≥ 1.13 and females aged between 64 and 83 years were predicted as having osteopenia. The decision tree's accuracy in the training set was 0.7627 (95 % CI (0.6341, 0.8638)), and its accuracy in the test set was 0.7895 (95 % CI (0.6268, 0.9045)). Multinomial logistic regression results showed that humeral shaft medullary cavity ratios M2/M4, age, and gender in X-ray images were significantly associated with the occurrence of osteoporosis.

Conclusion

Utilizing X-ray data of the proximal humerus in conjunction with demographic information such as gender and age enable the prediction of localized osteoporosis, facilitating physicians' rapid comprehension of osteoporosis in patients and optimization of clinical treatment plans.

Level of evidence

Level IV retrospective case study.

Association of apolipoprotein A1 levels with lumbar bone mineral density and β-CTX in osteoporotic fracture individuals: a cross-sectional investigation

Background

The relationship between the levels of high-density lipoprotein (HDL) and bone mineral density (BMD) is controversial. Furthermore, the specific role of apolipoprotein A1 (APOA1), a primary HDL component, in regulating BMD remains unclear. This study aimed to elucidate the correlation between APOA1 levels and lumbar BMD in patients with osteoporotic fracture (OPF) for novel insights into potential therapeutic strategies against osteoporosis.

Methods

This study included 587 OPF patients enrolled at the Kunshan Hospital, Affiliated with Jiangsu University between January 2017 and July 2022. The patient's serum APOA1 levels were determined, followed by the assessment of lumbar BMD and C-terminal telopeptide of type I collagen (β-CTX) as outcome variables. The association of APOA1 levels with lumbar BMD and β-CTX was assessed via Generalized Estimating Equations (GEE) and spline smoothing plot analyses. A generalized additive model (GAM) helped ascertain non-linear correlations. Moreover, a subgroup analysis was also conducted to validate the result's stability.

Results

It was observed that APOA1 levels were positively correlated with lumbar BMD (β = 0.07, 95% CI: 0.02 to 0.11, p = 0.0045), indicating that increased APOA1 levels were linked with enhanced lumbar BMD. Furthermore, APOA1 levels were negatively related to β-CTX (β = -0.19, 95% CI: -0.29 to -0.09, p = 0.0003), suggesting APOA1 might reduce osteolysis. In addition, these findings were robustly supported by subgroup and threshold effect analyses.

Conclusion

This study indicated that increased APOA1 levels were correlated with enhanced lumbar BMD and decreased osteolysis in OPF patients. Therefore, APOA1 may inhibit osteoclast activity to prevent further deterioration in osteoporotic patients. However, further research I warranted to validate these conclusions and elucidate the underlying physiologies.

Using machine learning techniques to predict the risk of osteoporosis based on nationwide chronic disease data

Osteoporosis is a major public health concern that significantly increases the risk of fractures. The aim of this study was to develop a Machine Learning based predictive model to screen individuals at high risk of osteoporosis based on chronic disease data, thus facilitating early detection and personalized management. A total of 10,000 complete patient records of primary healthcare data in the German Disease Analyzer database (IMS HEALTH) were included, of which 1293 diagnosed with osteoporosis and 8707 without the condition. The demographic characteristics and chronic disease data, including age, gender, lipid disorder, cancer, COPD, hypertension, heart failure, CHD, diabetes, chronic kidney disease, and stroke were collected from electronic health records. Ten different machine learning algorithms were employed to construct the predictive mode. The performance of the model was further validated and the relative importance of features in the model was analyzed. Out of the ten machine learning algorithms, the Stacker model based on Logistic Regression, AdaBoost Classifier, and Gradient Boosting Classifier demonstrated superior performance. The Stacker model demonstrated excellent performance through ten-fold cross-validation on the training set and ROC curve analysis on the test set. The confusion matrix, lift curve and calibration curves indicated that the Stacker model had optimal clinical utility. Further analysis on feature importance highlighted age, gender, lipid metabolism disorders, cancer, and COPD as the top five influential variables. In this study, a predictive model for osteoporosis based on chronic disease data was developed using machine learning. The model shows great potential in early detection and risk stratification of osteoporosis, ultimately facilitating personalized prevention and management strategies.

Applications of Artificial Intelligence Methods for the Prediction of Osteoporotic Fractures

Background: Osteoporosis is a socio-economic problem of modern aging societies. Bone fractures and the related treatments generate the highest costs. The occurrence of osteoporotic fractures is a cause of chronic disability, many complications, reduced quality of life, and often premature death. Aim of the study: The aim of the study was to determine which of the patient's potential risk factors pertaining to various diseases and lifestyle have an essential impact on the occurrence of low-energy fractures and the hierarchy of these factors. Methods: The study was retrospective. The documentation of 222 patients (206 women and 16 men) from an osteoporosis treatment clinic in Łódź, Poland was analyzed. Each patient was described by a vector consisting of 27 features, where each feature was a different risk factor. Using artificial neural networks, an attempt was made to create a model that, based on the available data, would be able to predict whether the patient would be exposed to low-energy fractures. We developed a neural network model that achieved the best result for the testing data. In addition, we used other methods to solve the classification problem, i.e., correctly dividing patients into two groups: those with fractures and those without fractures. These methods were logistic regression, k-nearest neighbors and SVM. Results: The obtained results gave us the opportunity to assess the effectiveness of various methods and the importance of the features describing patients. Using logistic regression and the recursive elimination of features, a ranking of risk factors was obtained in which the most important were age, chronic kidney disease, neck T-score, and serum phosphate level. Then, we repeated the learning procedure of the neural network considering only these four most important features. The average mean squared error on the test set was about 27% for the best variant of the model. Conclusions: The comparison of the rankings with different numbers of patients shows that the applied method is very sensitive to changes in the considered data (adding new patients significantly changes the result). Further cohort studies with more patients and more advanced methods of machine learning may be needed to identify other significant risk factors and to develop a reliable fracture risk system. The obtained results may contribute to the improved identification patients at risk of low-energy fractures and early implementation of comprehensive treatment.