Automated Sulcus Depth Measurement on Axial Knee MR Images

Abstract

Patellofemoral instability is a knee disorder in which the patella, slips out of its usual placement leading to knee pain. The patella may displace from its position due to abnormality in the shape of patellar surface on femur bone. The orthopaedic experts manually measure certain parameters from the available axial knee scans for the patellar instability diagnosis, which is labor-intensive and susceptible to inter-and intra-observer variations. The automated segmentation of femur region in knee magnetic resonance (MR) image can help in easily identifying the abnormality in the patellar surface. Therefore, in this paper, the femur bone in the axial knee MR scans has been segmented using two variants of U-Net: basic U-Net and U-Net++ and the results have been compared and visualized. The validation dice similarity coefficient (DSC) and accuracy of 89.62% and 94.05% were obtained, respectively, for U-Net. For U-Net++, the validation DSC of 95.04% and validation accuracy of 94.91% were obtained. Further, in this paper, the sulcus depth measurement has been automated using basic image processing techniques. A mean error of 0.565 mm was obtained when tested on 20 axial knee MR images. The T2-weighted knee MRI dataset of 55 patients has been acquired from Post Graduate Institute of Medical Education and Research (PGIMER), Chandigarh for training and testing of the proposed approach.