Computer-Aided Diagnosis of Vertebral Compression Fractures Using Convolutional Neural Networks and Radiomics.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Orthopedics

Background:

• Vertebral Compression Fractures (VCFs) require accurate etiological classification for appropriate treatment and prognosis.
• Non-traumatic VCFs are categorized as benign (osteoporosis-related) or malignant (tumor-related).
• Current diagnostic methods face challenges with limited data availability for deep learning models and underutilization of supplementary clinical information.

Purpose of the Study:

• To develop and evaluate a hybrid method for classifying non-traumatic VCFs.
• To integrate features from Convolutional Neural Networks (CNNs), radiomics, and clinical data for improved classification accuracy.
• To leverage a Genetic Algorithm for feature selection and CNN hyper-parameter optimization.

Main Methods:

• A hybrid classification model was proposed, combining intermediate CNN features, radiomic features, and clinical/histogram data.
• External features were incorporated as additional inputs to the first dense layer of a CNN.
• A Genetic Algorithm was employed for selecting relevant features and optimizing CNN hyperparameters.

Main Results:

• Experimental results demonstrated that combining diverse data sources significantly improves VCF classification performance.
• Pre-trained CNN models outperformed models trained from scratch, indicating the benefit of transfer learning.
• The hybrid approach effectively integrates heterogeneous data for enhanced diagnostic capabilities.

Conclusions:

• The proposed hybrid method offers a promising approach for accurate VCF classification, outperforming traditional methods.
• Integrating radiomic, clinical, and deep learning features enhances diagnostic precision in VCF etiology determination.
• The study highlights the potential of hybrid AI models in medical image analysis, particularly for conditions with limited data.