Performance of the deep convolutional neural network based magnetic resonance image scoring algorithm for...

Area of Science:

• Radiology
• Artificial Intelligence
• Infectious Diseases

Background:

• Differentiating tuberculous spondylitis from pyogenic spondylitis on MR imaging is crucial for appropriate treatment.
• Current diagnostic methods rely on expert radiologist interpretation, which can be subjective.

Purpose of the Study:

• To evaluate the diagnostic performance of a deep convolutional neural network (DCNN) in differentiating tuberculous spondylitis from pyogenic spondylitis using MR imaging.
• To compare the DCNN's performance against that of experienced musculoskeletal radiologists.

Main Methods:

• A retrospective study analyzed spine MR images from 161 patients (80 with tuberculous spondylitis, 81 with pyogenic spondylitis).
• A DCNN classifier was trained and validated using four-fold cross-validation.
• Three musculoskeletal radiologists independently interpreted the MR images.

Main Results:

• The DCNN achieved an area under the receiver operating characteristic curve (AUC) of 0.802.
• The pooled AUC for the three radiologists was 0.729.
• There was no statistically significant difference in diagnostic performance between the DCNN and the radiologists (P=0.079).

Conclusions:

• The DCNN demonstrates comparable diagnostic performance to skilled radiologists in differentiating tuberculous from pyogenic spondylitis on MR imaging.
• AI-powered tools like DCNNs show potential for assisting in the diagnosis of spinal infections.