Automated measurement and grading of knee cartilage thickness: a deep learning-based approach

• DOI: 10.3389/fmed.2024.1337993
• Publication Date: 2024-02-29T15:35:27Z
• Authors: JiangRong Guo; Pengfei Yan; Yong Qin; MeiNa Liu; Yingkai Ma; JiangQi Li; Ren Wang; Hao Luo; Songcen Lv
• Keywords: Knee cartilage; Grading (engineering)

Background Knee cartilage is the most crucial structure in knee, and reduction of thickness a significant factor occurrence development osteoarthritis. Measuring allows for more accurate assessment wear, but this process relatively time-consuming. Our objectives encompass using various DL methods to segment knee from MRIs taken with different equipment parameters, building DL-based model measuring grading cartilage, establishing standardized database thickness. Methods In retrospective study, we selected mixed MRI dataset consisting 700 cases four datasets varying We employed convolutional neural networks—UNet, UNet++, ResUNet, TransUNet—to train dataset, leveraging an extensive array labeled data effective supervised learning. Subsequently, measured graded 12 regions. Finally, standard was established 291 ages ranging 20 45 years Kellgren–Lawrence 0. Results The validation results network segmentation showed that TransUNet performed best overall dice similarity coefficient 0.813 Intersection over Union 0.692. model’s mean absolute percentage error automatic measurement after 0.831. experiment also yielded thickness, average 1.98 mm femoral 2.14 tibial cartilage. Conclusion By selecting network, built stronger generalization ability automatically segment, measure, grade This can assist surgeons accurately efficiently diagnosing changes patients’