Development and application of AI assisted automatic reconstruction of axial lumbar disc CT images and diagnosis of lumbar disc herniation.

RATIONALE AND OBJECTIVES

To evaluate the value of artificial intelligence (AI) assisted diagnostic system in reconstructing axial lumbar disc CT images and diagnosing lumbar disc herniation.

MATERIALS AND METHODS

440 patients with lumbar disc herniation were included, with 400 cases of spiral data (320 training, 40 validations, and 40 testing) and 40 cases of axial data (testing). V-Net was used to reconstruct the axial lumbar disc images. U-Net was used to segment the herniated discs and perform MSU classification. The Dice coefficient was used to evaluate the accuracy of AI in lumbar vertebras and discs segmentation. The quality of axial CT images reconstructed by AI and radiology technician was compared. The diagnostic accuracy of AI, radiologist, and AI + radiologist for the MSU classification of lumbar disc herniation in spiral and axial data was evaluated.

RESULTS

The Dice coefficients of AI for segmenting the sacral, lumbar, and lumbar discs were 0.953, 0.940, and 0.926, respectively. The quality of the axial CT images reconstructed by AI and radiographer had non-significant difference (P＞0.05). In both the spiral and axial data, the accuracy of AI, radiologist, and AI + radiologist in diagnosing the MSU classification was significantly different (P < 0.01). The diagnostic accuracy of the AI system in MSU classification was higher in the spiral data than that of the axial data (P = 0.003).

CONCLUSION

The AI system is feasible and satisfactory for segmentation of lumbar CT image, reconstruction of axial lumbar disc CT images, and diagnosis of lumbar disc herniation.