Comparing image normalization techniques in an end-to-end model for automated modic changes classification from MRI...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Modic Changes (MCs) are MRI signal alterations in vertebrae.
• Accurate MC assessment is crucial for spinal diagnostics and treatment planning.
• Current manual assessment can be time-consuming and subjective.

Purpose of the Study:

• To develop an end-to-end automated model for detecting and classifying Modic Changes (MCs) in lumbar MRIs.
• To evaluate the impact of MRI normalization techniques on MC classification accuracy.
• To assess the clinical applicability of the automated MC assessment model.

Main Methods:

• An integrated model combining Faster R-CNN for region detection and a 3D Convolutional Neural Network (CNN) for classification was utilized.
• The model processed paired T1- and T2-weighted lumbar MRI images.
• Two datasets were employed for model training and validation.

Main Results:

• The detection component achieved high accuracy in identifying intervertebral regions (IoU > 0.7).
• The classification task demonstrated superior performance with MRI standardization, yielding high sensitivity and balanced accuracy (0.80 F1 score).
• Specific mean sensitivities were 0.83 for MC0, 0.85 for MC1, and 0.78 for MC2.

Conclusions:

• The end-to-end automated model offers a promising approach for efficient and standardized Modic Change assessment in clinical practice.
• Standardization techniques significantly improve classification performance.
• Further research is needed for external validation and enhanced model generalization.