Differentiation of tuberculous and brucellar spondylitis using conventional MRI-based deep learning algorithms

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Tuberculous spondylitis (TS) and brucellar spondylitis (BS) are common spinal infections that can present with similar imaging findings on conventional MRI.
• Accurate differentiation is crucial for appropriate treatment and patient outcomes.

Purpose of the Study:

• To evaluate the feasibility of deep learning (DL) models in distinguishing between TS and BS using conventional MRI sequences.
• To compare the diagnostic performance of DL models with that of experienced radiologists.

Main Methods:

• A dataset of 383 patients with confirmed TS (n=182) or BS (n=201) was retrospectively collected.
• Sagittal T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and fat-suppressed T2WI were used to train and validate DL models (VGG19, VGG16, ResNet18, DenseNet121).
• Classification performance was assessed using the area under the receiver operating characteristic curve (AUC), and DL model performance was compared against two radiologists.

Main Results:

• Deep learning models achieved high AUC values, ranging from 0.801 to 0.973.
• The VGG19-based models demonstrated superior performance, with the combined T1WI, T2WI, and FS T2WI model achieving an optimal AUC of 0.973.
• All combined DL models significantly outperformed the diagnostic accuracy of the two radiologists (P<0.05).

Conclusions:

• Deep learning models show significant potential for aiding in the differentiation of tuberculous spondylitis and brucellar spondylitis on conventional MRI.
• These AI-driven tools can serve as a valuable reference for clinicians in the diagnosis of spinal infections.