Osteoporosis is a common skeletal disorder characterized by reduced bone density and deterioration of bone microarchitecture, increasing the risk of fractures and serious health complications. Osteoporosis can progress without obvious symptoms until a fracture occurs, making early detection and accurate assessment of bone health critically important.
An important stage in the progression of bone loss is Osteopenia, an intermediate condition in which bone density is lower than normal but has not yet progressed to full osteoporosis. Identifying osteopenia at an early stage can provide an important opportunity for clinical intervention and prevention before significant bone deterioration and fractures occur.
Traditional osteoporosis screening commonly relies on Dual-energy X-ray Absorptiometry (DEXA/DXA), while other medical imaging modalities, including X-ray, CT, and MRI, may also provide information related to bone health. However, limitations related to accessibility, cost, radiation exposure in certain imaging modalities, and the difficulty of identifying subtle changes in bone structure highlight the potential value of computer-aided diagnostic technologies.
This research presentation, produced as part of the scientific research productions of Knowture, explores the development of a Computer-Aided Diagnosis (CAD) System for screening osteoporosis and its early stage, osteopenia, using standard knee X-ray images and Artificial Intelligence techniques.
The research was presented by Assoc. Prof. Fatma El-Zahraa Ahmed, Assistant Professor at the Information Technology Department, Faculty of Computers and Information, Mansoura University. The research work was conducted by Hadeel Osama Elsisi, Information Technology Department, Faculty of Computers and Information, Mansoura University, under the supervision of Prof. Noha Ahmed Hikal and Assoc. Prof. Fatma El-Zahraa Ahmed.
Study Overview
The research focuses on developing an intelligent computer-aided diagnostic framework capable of classifying knee X-ray images into three clinically relevant categories:
- Normal Bone
- Osteopenia
- Osteoporosis
Rather than limiting the diagnosis to a simple presence-or-absence classification, the proposed approach aims to recognize different stages of bone health and provide probabilistic severity scores that reflect the continuous progression of bone loss.
This approach is particularly important for osteopenia, which represents an early and potentially actionable stage of bone deterioration. According to the presentation, existing artificial intelligence models can face difficulties distinguishing subtle visual differences between bone-health stages, particularly when dealing with the visual ambiguity of bone textures.
The proposed framework therefore focuses on extracting fine-grained visual characteristics from knee X-ray images and combining deep-learning-based feature extraction with machine learning classification to support accurate multi-stage diagnosis.
Research Motivation
The motivation behind this research is closely connected to the silent progression of osteoporosis and the importance of identifying bone deterioration before severe fractures occur.
Late detection of osteoporosis can result in serious complications, including painful fractures, reduced mobility, and potentially life-threatening consequences following major fractures. The presentation highlights the need for accurate tools capable of detecting and staging bone health at an earlier point in the disease process.
The research addresses several important challenges:
Silent Disease Progression
Osteoporosis can develop without noticeable symptoms until significant bone loss has already occurred.
Limitations of Conventional Screening
Advanced diagnostic techniques such as DEXA can be costly or less accessible in some settings, creating interest in more widely available imaging-based screening approaches.
Difficulty Detecting Osteopenia
The visual differences between normal bone, osteopenia, and osteoporosis can be subtle, making early-stage classification particularly challenging for both conventional analysis and some existing AI systems.
Need for More Informative Classification
Instead of using only hard diagnostic labels, the proposed framework investigates probabilistic severity assessment to provide a more nuanced representation of the patient's bone-health stage.
Proposed Computer-Aided Diagnosis Framework
The research develops an AI-based framework using knee X-ray images as the primary imaging source.
The dataset used in the research contains knee X-ray scans classified into Normal, Osteopenia, and Osteoporosis, together with demographic and clinical information such as age, gender, medical history, and fracture history.
The proposed processing pipeline includes several key stages.
Image Preprocessing
The X-ray images undergo preprocessing procedures including:
- Grayscale conversion.
- Image resizing to 660 × 660 pixels.
- Data augmentation using geometric and photometric transformations.
- Pixel normalization.
- Dataset balancing to reduce potential model bias.
These preprocessing steps are designed to improve data consistency, preserve relevant bone-texture information, and enhance the robustness of the classification system.
Deep Feature Extraction Using VGG16
The framework uses VGG16 to extract discriminative spatial features from knee X-ray images.
The architecture's stacked 3×3 convolutional filters are used to capture detailed local characteristics and subtle textural patterns that may be relevant to identifying early-stage osteopenia.
Classification Using SVM
The extracted features are subsequently classified using a Support Vector Machine (SVM) with an RBF kernel.
The RBF-based classifier is designed to handle nonlinear relationships between the different bone-health categories and to separate complex, overlapping classes. The framework also uses calibrated probabilities to estimate the severity of bone loss rather than relying exclusively on a hard classification label.
Experimental Results
The proposed framework demonstrated strong classification performance across the three studied categories: Normal, Osteopenia, and Osteoporosis.
The presentation reports an overall accuracy of 94.85% for the proposed VGG16 + SVM (RBF) framework when classifying the three stages. The reported class-level results include:
- Normal: Precision 100%, Recall 94%, F1-Score 0.97
- Osteopenia: Precision 89%, Recall 100%, F1-Score 0.94
- Osteoporosis: Precision 98%, Recall 90%, F1-Score 0.94
The reported ROC analysis also achieved an AUC of 0.99, demonstrating strong discrimination among the classes studied.
The framework was further evaluated using cross-validation. The presentation reports average validation accuracies of:
- 3-Fold: 92.67%
- 5-Fold: 93.60%
- 7-Fold: 93.19%
- 10-Fold: 93.40%
These results were presented as evidence supporting the robustness of the framework and the absence of significant overfitting.
Comparison with Other Machine Learning Classifiers
The proposed VGG16 + SVM (RBF) framework was compared with several conventional machine learning classifiers using the same knee X-ray dataset.
The reported overall accuracies were:
- MLP: 91.75%
- KNN: 93.20%
- Decision Tree: 89.17%
- Logistic Regression: 91.23%
- SVM with RBF kernel: 94.85%
The results indicate that the proposed SVM-based approach achieved the highest overall accuracy among the classifiers evaluated in the study.
The presentation also compares the proposed framework with selected previous approaches and reports higher accuracy for the proposed VGG16 + SVM framework in both binary and three-class diagnostic settings.
Disease Severity Assessment
One of the notable aspects of the research is its focus on moving beyond conventional categorical classification toward probabilistic disease-severity assessment.
The framework can provide probability distributions across the three studied stages, allowing the system to indicate how strongly an X-ray corresponds to Normal, Osteopenia, or Osteoporosis.
This approach may provide a more informative representation of bone-health status, particularly for cases located near the transition between normal bone density and osteoporosis.
The presentation describes this capability as a “Disease Severity” metric, designed to support more personalized clinical assessment and potentially assist in treatment planning.
What You Will Learn in This Video
This research presentation provides an overview of how Artificial Intelligence and Machine Learning can be applied to medical image analysis for osteoporosis screening.
By watching the presentation, viewers will gain insights into:
- The clinical significance of osteoporosis and osteopenia.
- The progression from normal bone health to osteopenia and osteoporosis.
- Current approaches to osteoporosis diagnosis and screening.
- The role of Computer-Aided Diagnosis (CAD) systems in medical imaging.
- The challenges are associated with detecting early-stage osteopenia.
- The use of knee X-ray images for AI-assisted osteoporosis screening.
- Image preprocessing and data augmentation techniques.
- Feature extraction using the VGG16 deep learning architecture.
- Classification using an SVM with an RBF kernel.
- Multi-class classification of Normal, Osteopenia, and Osteoporosis.
- Probabilistic disease-severity assessment.
- Evaluation using accuracy, precision, recall, F1-score, ROC/AUC, and cross-validation.
- The potential role of AI in supporting earlier intervention and personalized clinical assessment.
Key Findings
High Diagnostic Accuracy
The proposed VGG16 + SVM (RBF) framework achieved an overall accuracy of 94.85% for three-class classification of Normal, Osteopenia, and Osteoporosis.
Strong Discriminative Performance
The ROC analysis reported an AUC of 0.99, indicating strong classification performance across the three categories studied.
Effective Osteopenia Detection
The framework achieved a reported 100% recall for the Osteopenia class, an important finding given the research's focus on identifying this early stage of bone deterioration.
Multi-Stage Diagnosis
The proposed system distinguishes between Normal, Osteopenia, and Osteoporosis rather than treating osteoporosis as a simple binary classification problem.
Probabilistic Severity Assessment
The research introduces a disease-severity perspective that provides probabilistic information about the patient's stage of bone health, potentially offering more nuanced diagnostic insight.
Why This Research Matters
The significance of this research extends beyond achieving high classification accuracy.
Osteoporosis represents a major public-health challenge because bone deterioration can progress silently until a serious fracture occurs. Identifying osteopenia before the condition develops into full osteoporosis may provide an important window for prevention, monitoring, and early intervention.
The use of standard knee X-ray images is particularly relevant because X-ray imaging is widely available compared with some advanced diagnostic technologies. Developing AI systems capable of extracting useful information from routinely acquired images could potentially contribute to more accessible screening strategies.
The research also demonstrates the value of combining Deep Learning feature extraction with traditional machine learning classification. Rather than relying solely on an end-to-end deep learning model, the proposed approach uses VGG16 to capture detailed image features and an RBF-based SVM to perform nonlinear classification.
Future directions identified in the presentation include synthetic X-ray data generation using GANs, multimodal fusion incorporating patient metadata such as age and BMI, and the implementation of Explainable AI techniques using visual heatmaps to help clinicians understand model predictions.
Who Should Watch This Research Presentation?
This research presentation is particularly valuable for:
- Radiologists and medical imaging specialists are interested in AI-assisted osteoporosis screening.
- Orthopedic physicians and healthcare professionals are involved in bone-health assessment.
- Researchers working in Artificial Intelligence and Medical Image Analysis.
- Computer Science and Information Technology researchers developing healthcare applications.
- Biomedical and healthcare technology professionals.
- Postgraduate students studying Machine Learning, Deep Learning, Computer Vision, and Medical Informatics.
- Researchers are interested in Computer-Aided Diagnosis (CAD) systems.
- Healthcare professionals are interested in the application of AI for early disease detection.
About Research
Researcher:
Hadeel Osama Elsisi
Information Technology Department, Faculty of Computers and Information, Mansoura University.
Presented by:
Assoc. Prof. Fatma El-Zahraa Ahmed
Assistant Professor, Information Technology Department, Faculty of Computers and Information, Mansoura University.
Supervision:
Prof. Noha Ahmed Hikal
Dean, Faculty of Computers and Information, Mansoura University.
Assoc. Prof. Fatma El-Zahraa Ahmed
Information Technology Department, Faculty of Computers and Information, Mansoura University.
Publication Reference
Research Title:
Computer-Aided Diagnosis System for Screening Osteoporosis and its Early Stage (Osteopenia)
Researcher: Hadeel Osama Elsisi
Department: Information Technology Department
Faculty: Faculty of Computers and Information, Mansoura University
Research Link:
(رابط البحث)
Note: Publication details such as journal name, volume, issue, pages, publication date, and DOI have not been added here because they are not provided in the presentation. They can be inserted once the official publication information is available.
Related Research References
The presentation discusses and compares the proposed framework with previous research in AI-based osteoporosis detection, including studies using CNNs, Vision Transformers, ensemble learning, and knee X-ray imaging.
Selected references cited in the presentation include:
- Kumar A, Joshi RC, Dutta MK, Burget R, Myska V. Osteo-Net: A Robust Deep Learning-Based Diagnosis of Osteoporosis Using X-ray Images. 2022 International Conference on Telecommunications and Signal Processing (TSP).
- Dodamani S, et al. Vision Transformer for Bone Mineral Density Classification. Multimedia Tools and Applications, 2022.
- Qureshi MB, et al. Deep-learning based osteoporosis classification in knee X-rays using transfer-learning approach. Scientific Reports, 2025.
- Narayanan S, et al. WKFF-Net: A Weighted Kernel Feature Fusion Network for Bone Density Estimation from Knee X-rays. Computerized Medical Imaging and Graphics, 2026.
- Wani IM, Arora S. Osteoporosis diagnosis in knee X-rays by transfer learning based on Convolutional Neural Network. Multimedia Tools and Applications, 2023.
About Knowture Platform
Knowture is the academic digital platform of the Postgraduate Studies and Research Sector at Mansoura University.
Through research presentations, expert lectures, educational programs, and evidence-based scientific content, Knowture aims to transform academic knowledge and research achievements into accessible digital learning experiences for postgraduate students, researchers, academics, and professionals worldwide.
The Knowture Research Video Library provides a dedicated space for showcasing scientific research and highlighting innovative applications of technology, Artificial Intelligence, and interdisciplinary research.
