The capabilities of artificial intelligence in segmentation and detection of spinal images at the current stage of development: a systematic review

Abstract

INTRODUCTION : From the very beginning of the theoretical substantiation of the creation of artificial intelligence, there were ideas about creating an electronic assistant for a doctor. The most effective for solving such a problem in medical diagnostics are complex systems based on one of the most advanced methods of artificial intelligence — neural networks. Study design: a systematic review. OBJECTIVE : Determining the capabilities of neural networks at the current stage of their development in the field of such tasks as segmentation and detection of spine images. MATERIALS AND METHODS : Using the PRISMA protocol, a search was performed in the Pubmed database for the period from January 2017 to December 31, 2022 using keywords «deep learning» OR «neural network» OR «artificial Intelligence», AND («spine») AND («detection» OR «segmentation»). RESULTS : For the systematic review, 30 articles were selected that described such functions of neural networks as segmentation and detection in the analysis of spinal images. DISCUSSION: Based on the analysis of literature sources, conclusions were made about the usefulness of using artificial intelligence at the current stage of development in vertebrology. CONCLUSION : Neural networks cope quite well with the tasks of segmentation and detection in vertebrology. In segmentation tasks, there is a clear leader — U-Net and its various modifications. In detection, the leading architectures are: SCRL, 3D FCN, CNN of our own design and a combination of Mask R-CNN+ResNet101 networks. Software based on the work of artificial intelligence can help both a radiologist and a vertebrologist reduce the load and simplify the work by automating and semi-automating the diagnostic process.

Ähnliche Arbeiten