ObjectiveIn recent years, artificial intelligence deep learning has gradually become a research hotspot in diagnosing and treating cervical and lumbar degenerative diseases, represented by cervical spondylosis and lumbar disc herniation. MethodsRetrospective inclusion of 590 patients with lumbar disc herniation, randomly divided into a training group (n = 456) and a testing group (n = 134), labeled with intervertebral disc herniation areas on lumbar MR images (1157 in the training group and 395 in the testing group). Construct a deep learning model based on Unet to segment and detect intervertebral disc herniation areas in sagittal and transverse planes and evaluate the model's lesion detection efficiency and accuracy in lesion segmentation. ResultsThe model has good detection performance on the test group data. The F1 values in the sagittal and transverse positions are 0.971 (95% CI: 0.951–0.987) and 0.903 (95% CI: 0.870–0.935). In the segmentation of prominent areas, the Dice values in the sagittal and transverse planes are 0.774 ± 0.193 and 0.647 ± 0.261 (2D plane), 0.777 ± 0.148 and 0.655 ± 0.180 (3D plane), respectively. ConclusionThe deep learning model based on U-net can accurately identify and segment the intervertebral disc herniation area in lumbar MR images, which helps to assist clinical diagnosis.
Read full abstract