Auto-segmentation of the tibia and femur from knee MR images via deep learning and its application to cartilage strain and recovery.

The ability to efficiently and reproducibly generate subject-specific 3D models of bone and soft tissue is important to many areas of musculoskeletal research. However, methodologies requiring such models have largely been limited by lengthy manual segmentation times. Recently, machine learning, and more specifically, convolutional neural networks, have shown potential to alleviate this bottleneck in research throughput. Thus, the purpose of this work was to develop a modified version of the convolutional neural network architecture U-Net to automate segmentation of the tibia and femur from double echo steady state knee magnetic resonance (MR) images. Our model was trained on a dataset of over 4,000 MR images from 34 subjects, segmented by three experienced researchers, and reviewed by a musculoskeletal radiologist. For our validation and testing sets, we achieved dice coefficients of 0.985 and 0.984, respectively. As further testing, we applied our trained model to a prior study of tibial cartilage strain and recovery. In this analysis, across all subjects, there were no statistically significant differences in cartilage strain between the machine learning and ground truth bone models, with a mean difference of 0.2 ± 0.7 % (mean ± 95 % confidence interval). This difference is within the measurement resolution of previous cartilage strain studies from our lab using manual segmentation. In summary, we successfully trained, validated, and tested a machine learning model capable of segmenting MR images of the knee, achieving results that are comparable to trained human segmenters.

The Predicted Position of the Knee Near the Time of ACL Rupture Is Similar Between 2 Commonly Observed Patterns of Bone Bruising on MRI.

BACKGROUND: Bone bruises observed on magnetic resonance imaging (MRI) can provide insight into the mechanisms of noncontact anterior cruciate ligament (ACL) injury. However, it remains unclear whether the position of the knee near the time of injury differs between patients evaluated with different patterns of bone bruising, particularly with regard to valgus angles. HYPOTHESIS: The position of the knee near the time of injury is similar between patients evaluated with 2 commonly occurring patterns of bone bruising. STUDY DESIGN: Descriptive laboratory study. METHODS: Clinical T2- and T1-weighted MRI scans obtained within 6 weeks of noncontact ACL rupture were reviewed. Patients had either 3 (n = 20) or 4 (n = 30) bone bruises. Patients in the 4-bone bruise group had bruising of the medial and lateral compartments of the femur and tibia, whereas patients in the 3-bone bruise group did not have a bruise on the medial femoral condyle. The outer contours of the bones and associated bruises were segmented from the MRI scans and used to create 3-dimensional surface models. For each patient, the position of the knee near the time of injury was predicted by moving the tibial model relative to the femoral model to maximize the overlap of the tibiofemoral bone bruises. Logistic regressions (adjusted for sex, age, and presence of medial collateral ligament injury) were used to assess relationships between predicted injury position (quantified in terms of knee flexion angle, valgus angle, internal rotation angle, and anterior tibial translation) and bone bruise group. RESULTS: The predicted injury position for patients in both groups involved a flexion angle <20°, anterior translation >20 mm, valgus angle <10°, and internal rotation angle <10°. The injury position for the 3-bone bruise group involved less flexion (odds ratio [OR], 0.914; 95% CI, 0.846-0.987; P = .02) and internal rotation (OR, 0.832; 95% CI, 0.739-0.937; P = .002) as compared with patients with 4 bone bruises. CONCLUSION: The predicted position of injury for patients displaying both 3 and 4 bone bruises involved substantial anterior tibial translation (>20 mm), with the knee in a straight position in both the sagittal (<20°) and the coronal (<10°) planes. CLINICAL RELEVANCE: Landing on a straight knee with subsequent anterior tibial translation is a potential mechanism of noncontact ACL injury.

Distribution of Bone Contusion Patterns in Acute Noncontact Anterior Cruciate Ligament-Torn Knees.

BACKGROUND: Bone contusions are commonly observed on magnetic resonance imaging (MRI) in individuals who have sustained a noncontact anterior cruciate ligament (ACL) injury. Time from injury to image acquisition affects the ability to visualize these bone contusions, as contusions resolve with time. PURPOSE: To quantify the number of bone contusions and their locations (lateral tibial plateau [LTP], lateral femoral condyle [LFC], medial tibial plateau [MTP], and medial femoral condyle [MFC]) observed on MRI scans of noncontact ACL-injured knees acquired within 6 weeks of injury. STUDY DESIGN: Cross-sectional study; Level of evidence, 3. METHODS: We retrospectively reviewed clinic notes, operative notes, and imaging of 136 patients undergoing ACL reconstruction. The following exclusion criteria were applied: MRI scans acquired beyond 6 weeks after injury, contact ACL injury, and previous knee trauma. Fat-suppressed fast spin-echo T2-weighted MRI scans were reviewed by a blinded musculoskeletal radiologist. The number of contusions and their locations (LTP, LFC, MTP, and MFC) were recorded. RESULTS: Contusions were observed in 135 of 136 patients. Eight patients (6%) had 1 contusion, 39 (29%) had 2, 41 (30%) had 3, and 47 (35%) had 4. The most common contusion patterns within each of these groups were 6 (75%) with LTP for 1 contusion, 29 (74%) with LTP/LFC for 2 contusions, 33 (80%) with LTP/LFC/MTP for 3 contusions, and 47 (100%) with LTP/LFC/MTP/MFC for 4 contusions. No sex differences were detected in contusion frequency in the 4 locations (P > .05). Among the participants, 50 (37%) had medial meniscal tears and 52 (38%) had lateral meniscal tears. CONCLUSION: The most common contusion patterns observed were 4 locations (LTP/LFC/MTP/MFC) and 3 locations (LTP/LFC/MTP).

Immune cell profiles in synovial fluid after anterior cruciate ligament and meniscus injuries.

BACKGROUND: Anterior cruciate ligament (ACL) and meniscus tears are common knee injuries. Despite the high rate of post-traumatic osteoarthritis (PTOA) following these injuries, the contributing factors remain unclear. In this study, we characterized the immune cell profiles of normal and injured joints at the time of ACL and meniscal surgeries. METHODS: Twenty-nine patients (14 meniscus-injured and 15 ACL-injured) undergoing ACL and/or meniscus surgery but with a normal contralateral knee were recruited. During surgery, synovial fluid was aspirated from both normal and injured knees. Synovial fluid cells were pelleted, washed, and stained with an antibody cocktail consisting of fluorescent antibodies for cell surface proteins. Analysis of immune cells in the synovial fluid was performed by polychromatic flow cytometry. A broad spectrum immune cell panel was used in the first 10 subjects. Based on these results, a T cell-specific panel was used in the subsequent 19 subjects. RESULTS: Using the broad spectrum immune cell panel, we detected significantly more total viable cells and CD3 T cells in the injured compared to the paired normal knees. In addition, there were significantly more injured knees with T cells above a 500-cell threshold. Within the injured knees, CD4 and CD8 T cells were able to be differentiated into subsets. The frequency of total CD4 T cells was significantly different among injury types, but no statistical differences were detected among CD4 and CD8 T cell subsets by injury type. CONCLUSIONS: Our findings provide foundational data showing that ACL and meniscus injuries induce an immune cell-rich microenvironment that consists primarily of T cells with multiple T helper phenotypes. Future studies investigating the relationship between immune cells and joint degeneration may provide an enhanced understanding of the pathophysiology of PTOA following joint injury.