Effects of noise on neural network based semantic segmentation of lumbar MRI for stenosis boundary delineation.

In this paper, we aim to provide a method based on deep learning to help clinicians diagnose lumbar spine MRI for stenosis more efficiently and with instruments that produce noisy images to make maximal usage of medical resources. To achieve the same, this paper evaluates the effect of various MRI noises, such as Gaussian, salt and pepper, speckled, etc. in boundary delineation of stenosis in MRIs of the lumbar spine using recently introduced semantic segmentation algorithm based on deep neural networks. A U-Net model was trained for segmentation and its performancewas tested on various MRI images with added noise. The proposed methodology adopted in this paper is successful in producing a considerably good performance with higher scores, which helps in identifying Lumbar Spinal Stenosis in a more efficient manner. [ABSTRACT FROM AUTHOR]