Friday, December 1, 2023

Association of AI-determined Kellgren–Lawrence grade with medial meniscus extrusion and cartilage thickness by AI-based 3D MRI analysis in early knee osteoarthritis – Scientific Reports

We determined the relationship between KL grade and MME and the anatomical relationship between KL grade and cartilage thickness in both early and moderate OA. The MME width was unchanged in KL grade 1, slightly increased in KL grade 2, and moderately increased in KL grade 3. Cartilage thinning in the medial femur was first noted in the anterior central subregion in KL grade 1, then it expanded inwardly in KL grade 2, and further expanded externally and into the middle subregions in KL grade 3. Cartilage thinning in the medial tibia first occurred in the anterior and middle external subregions in KL grade 1, expanded into the anterior and middle central subregion in KL grade 2, and expanded further into the posterior external subregion in KL grade 3.

The MME remained unchanged in KL grade 1, and the meniscal extrusion worsened from KL grade 2 and the grades thereafter. A positive correlation between MME and tibial osteophyte width on radiographs9 and MRI7 has been previously reported, as has a correlation between the KL grade and MME21. However, whether the MME differed between KL grade 0 and KL grade 1, or between KL grade 1 and KL grade 2 was not entirely evident, due to the low repeatability of KL evaluations in early to moderate OA19.

Whether the osteophytes or the MME occurs first in OA is unclear. The lack of a difference in MME between KL grades 0 and 1 in our study could suggest that osteophytes occur first. However, it is important to consider that all MRI scans in this study were conducted in the supine position, which could affect the measurement of MME, as MME has been reported to increase during weight-bearing scenarios22. This implies that the MRI-based measurements of MME in this study may be biased toward lower values than would be observed in a weight-bearing position. Similarly, cartilage thickness is known to decrease during weight-bearing23, particularly in central regions24. Therefore, the measurements of MME and cartilage thickness using MRI in this study could be influenced by the use of non-weight-bearing positions, lending further support for evaluation in the context of the potential bias introduced by the lack of weight-bearing, based on previous research. Consequently, due to the possible confounding effects of weight-bearing positioning, this study did not definitively determine whether osteophytes or MME occur first in OA.

Cartilage thinning in the medial tibia was first observed in the anterior and middle external subregions in KL grade 1, and it was subsequently observed in the surrounding areas. Our previous 3D MRI analysis of subjects with cartilage defects on the medial tibia revealed the presence of cartilage defects in an area that extended from the middle external subregion to the periphery when arranged according to the defect size8. Conversely, unlike those previous findings, thinning was observed in the present study in the anterior external subregion, as well as in the middle external subregion, in KL grade 1. This difference is important because KL grade 1 is considered to represent pre-radiographic OA and is not supposed to show any cartilage damage.

The femoral cartilage thinning initiated from the anterior central subregion and spread around the periphery. The anterior and middle subregions can be seen when viewed from below along the long axis of the femur, suggesting that these subregions are the areas where contact occurs during knee extension in the standing position and could explain the thinning of these subregions observed in KL grade 3. By contrast, the lack of thinning in the posterior subregions up to KL grade 3 and the thickening of cartilage in the posterior external subregion could reflect less loading on the posterior subregions25.

The observed differences in cartilage thickness between KL grade 0 and KL grade 1 are clinically relevant because they provide insight into the early stages of knee joint changes. The thinning of cartilage in KL grade 1 suggests that the disease process has already begun, and this could have implications for disease progression and treatment. In a clinical context, these findings could influence decisions regarding interventions and patient education. For example, individuals with KL grade 1 who may have unequivocal radiographic changes could still benefit from early interventions aimed at preserving joint function and preventing further progression of cartilage loss. These interventions could include recommendations for lifestyle changes, targeted exercise, and patient education about joint health26.

The association between MME width and cartilage thickness by KL grade in meMT revealed no association with KL grades 0 and 1, but a significant correlation with KL grade 2, and a stronger correlation with KL grade 3. KL grade 2 is characterized by unequivocal osteophytes, which increase the range of associated MME26, and this would reveal an association between MME width and cartilage thickness. KL grade 3 is characterized by joint space narrowing, which increases the range of cartilage thickness, and this would reveal a stronger association between MME width and cartilage thickness. A similar explanation could be proposed for mcMF.

The areas of cartilage thinning in each KL in this study suggested the following mechanism for OA progression, based on three assumptions: the tibial cartilage that is no longer covered by the meniscus is easily worn away27; femoral cartilage wears away in the area that contacts the area of tibial cartilage thinning28; and lateral thrust is an aggravating factor in medial OA29. For KL grade 1, MRI shows no MME, and this causes thinning of the aeMT and meMT. The acMF of this mirror lesion is also thinned. KL grade 2 shows a mild MME, and the aeMT and the other three subregions adjacent to the aeMT are thinned. In addition, femoral cartilage thinning occurs in the aiMF rather than the aeMF due to a lateral thrust. For KL grade 3, the MME becomes moderate, and the tibial cartilage thins to the peMT. The femoral cartilage thins in the six anterior and middle subregions, which are the mirror lesions of these tibial lesions. The three posterior regions are less affected by loading and do not thin, whereas the peMF thickens.

Cartilage thickness changes were analyzed by White et al.30, who performed a gross analysis of 46 cases of tibial cartilage resected during unilateral knee OA arthroplasty. Assessment of the medial central tibial cartilage in the center from anterior to posterior revealed a decrease in cartilage thickness in the anterior and middle tibial cartilage30. This previous study is considered to have targeted KL grades 3–4, and our results are consistent with KL grade 3, as the cartilage thickness was reduced in acMT and mcMT. Reichenbach et al.31 analyzed the relationship between KL grade and cartilage thickness in 948 knees from the Framingham OA Study. In that study, the subregions corresponding to our definition of mcMF showed significantly thinner cartilage thickness in KL 2–3 and 4, and the subregion corresponding to mcMT showed significantly thinner cartilage thickness in KL 2–3 in females and in KL 2–4 in males, compared to KL 0. These findings are consistent with our KL grades 3–4, but in some respects, they are inconsistent with KL grade 2. This study differed from ours in that the MRI was taken with 3D FLASH-water excitation sequences, and the segmentation was done manually and analyzed separately for females and males. Guillard et al.32 performed 3D MRI analysis of 8890 knees from the Osteoarthritis Initiative to quantify cartilage thickness in the central medial femur and central medial tibia for each KL grade. Cartilage thickness was unchanged between KL grade 0 and KL grade 2, and it decreased gradually in KL grades 3 and 4. These regions corresponded to our mcMF and mcMT subregions, and their results are consistent with our results for mcMF. The advantage of our study, compared to these previous studies, is that we analyzed the surrounding 8 subregions for the femur and tibia in addition to the central subregions, and we added the MME to our analyses.

KL grading was done automatically using KOALA AI software. Rough checks were performed by human eyes, checking for laterality errors due to poor x-ray conditions, but we prevented bias by making no manual changes. Three previous reports have evaluated the accuracy of KOALA software. Nehrer et al.4 analyzed 124 knee radiographs from the Osteoarthritis Initiative study and reported a 21% increase in KL grade concordance when KOALA software was used than when the software was not used. Smolle et al.33 reported that the concordance rate for KL grades by three certified orthopedic surgeons for 124 knee radiographs was twice as high with KOALA software than without KOALA software. Neubauer et al.34 analyzed 69 subjects in the MLKOA physical therapy trial and reported an increase in KL grade concordance when KOALA software was used than when KOALA software was not used. The greatest problem with KL grading is the high variability among raters, but this problem can be reduced with the use of automatic KL grading software, such as KOALA software, especially when the readers are not musculoskeletal radiologists or when they lack extensive experience in clinical trial readings.

In the present study, knee radiographs were taken in extension, with the knees in the upright position. This was done to follow the original method for KL grading. Kothari et al.35 reported that a mild knee flexion position contributes to reproducible joint space width. In fact, KOALA software uses radiographs taken at mild flexion using the Synaflexer as training data. While we have empirically confirmed that KOALA software can classify the KL grade from radiographs with knee extension without notable issues, one limitation to report is that the inter-image error is higher for radiographs taken with knee extension than for radiographs acquired with mild knee flexion.

We have reported on the accuracy of 3D MRI analysis by SYNAPSE 3D in two previous papers. First, we ran a validation test for our algorithm by randomly selecting 108 of the 113 subjects for training, and the other 5 subjects were used for a validation test by computing the Dice similarity coefficient (DSC)36. The mean DSC was 0.99 for the femoral bone, 0.98 for the tibial bone, 0.91 for the femoral cartilage, and 0.89 for the tibial cartilage17. A DSC of 1 indicates a perfect match, and 0 indicates a perfect mismatch. Our values were over 0.98 for bone and over 0.89 for cartilage, indicating a high degree of matching. Second, nine healthy volunteers underwent MRI twice in the same day, and the interscan measurement error (the absolute difference obtained by subtracting the second MRI from the first MRI) was calculated at each region and subregion from the data. The interscan measurement error for the cartilage thickness in the medial femoral region was 0.03 ± 0.02 mm (mean ± standard deviation), and the interscan measurement error for the nine subregions of the medial femoral cartilage region ranged from 0.04 ± 0.02 mm to 0.11 ± 0.10 mm. The interscan measurement error for the cartilage thickness in the medial tibial region was 0.03 ± 0.02 mm, and the interscan measurement error of the nine subregions of the medial femoral cartilage region ranged from 0.04 ± 0.03 mm to 0.11 ± 0.07 mm9. These interscan measurement errors seem to be sufficient to conduct this study.

The slice showing the maximum transverse diameter of the tibia in the coronal plane was selected for the MME measurements. However, this method may not always be accurate, as the shape of the tibia can vary from person to person, and the maximum transverse diameter may not always show the maximum MME width. Three-dimensional imaging can improve accuracy and reduce limitations in the MME measurements by providing a more complete view of the MM and the tibia.

The subjects in this study were significantly older as the KL grade increased, and the BMI was significantly higher for KL grades 2–3 than for KL grade 0. Both the MME37 and the cartilage thickness loss38 are associated with age and BMI; therefore, age and BMI should be considered potential confounding factors. However, we did not adjust for these factors in our analysis due to the decreasing number of subjects across the increasing KL grades, which culminated in a very small sample size for KL grade 3. Taking this limitation into account, future investigations are encouraged to control for age and BMI to deepen our understanding of their roles in the association between KL grade, MME, and cartilage thickness reduction.

Tibiofemoral knee OA comprises medial and lateral tibiofemoral OA; however, only medial OA was analyzed in the present study. The main reason for this limitation was that the anatomy, kinematics, and loading during movement differ considerably between the medial and lateral knee compartments of the knee39. Therefore, we were concerned that the analysis would be too complicated if we included both compartments. Furthermore, medial OA occurs more commonly than lateral OA39, and medial OA is more likely to result in surgery40. The medial compartment is also the focus of disease-modifying osteoarthritis disease (DMOAD) clinical trials. For these reasons, lateral OA was excluded from the present study, and therefore, the findings of this study cannot be generalized to patients with lateral tibiofemoral knee OA.

Our study has several other limitations. First, the KL grade was determined using an automated grading system to improve repeatability. However, this would not necessarily be consistent with expert consensus. Second, because we used an automated MRI analysis system that cannot yet measure MME, we had to measure the MME manually. The inter-rater repeatability of each measurement is reliable9, but the potential for measurement errors cannot be ignored. Third, among the 469 subjects included in this study, only 3 were graded as KL grade 4. We included the results for KL grade 4, but we excluded this grade from the discussion. Fourth, the subjects analyzed in this study are not representative of the general population; therefore, the results of the study must be interpreted with caution before they can be generalized.

We propose three future studies. First, our evaluation of MME used only a single slice from the MRI coronal sequence; therefore, we may not have captured the entire MME pathology. MME can potentially move maximally in the anterior, anteromedial, and posteromedial directions; hence, the next step should be to perform a more detailed analysis by assessing the direction of MME in two dimensions. Second, because the current study focused on individuals with no history of knee symptoms, KL grades 0–2 accounted for 97% of the cohort. In the next study, we will focus on groups with a high percentage of KL grades 3–4. Third, this analysis used cross-sectional data from the Kanagawa Knee Study, although this study also has longitudinal data collected at one-year intervals. The next step would be to use this longitudinal data set to further analyze the relationship between KL grades and cartilage thickness loss/ MME progression.

Source link

Related Articles

Leave a Reply

Stay Connected

- Advertisement -spot_img

Latest Articles

%d bloggers like this: