Automatic quantification and grading of hip bone marrow oedema in ankylosing spondylitis based on deep learning.

OBJECTIVE: This study has developed a new automatic algorithm for the quantificationy and grading of ankylosing spondylitis (AS)-hip arthritis with magnetic resonance imaging (MRI). METHODS: (1) This study designs a new segmentation network based on deep learning, and a classification network based on deep learning. (2) We train the segmentation model and classification model with the training data and validate the performance of the model. (3) The segmentation results of inflammation in MRI images were obtained and the hip joint was quantified using the segmentation results. RESULTS: A retrospective analysis was performed on 141 cases; 101 patients were included in the derived cohort and 40 in the validation cohort. In the derivation group, median percentage of bone marrow oedema (BME) for each grade was as follows: 36% for grade 1 (<15%), 42% for grade 2 (15-30%),and 22% for grade 3 (≥30%). The accuracy of 44 cases on 835 AS images was 85.7%. Our model made 31 correct decisions out of 40 AS test cases. This study showed that THE accuracy rate 85.7%. CONCLUSIONS: An automatic computer-based analysis of MRI has the potential of being a useful method for the diagnosis and grading of AS hip BME.