Magnetic resonance imaging (MRI) is the modality of choice for diagnosing meniscal tears with both high sensitivity and specificity (1, 2). Conventional morphological MRI relies on the assessment of surface integrity and sub-surface signal intensity as indicators of tissue defects, which is qualitative and has limitations to detect the meniscal composition changes before surface breakdown or small tears (3). Relaxation time-based quantitative MRI such as T1 and T2 mappings have been widely used to investigate the compositional tissue features beyond morphology and structure (4, 5).

Diffusion MRI (dMRI) has been extensively used to reveal the microstructure of different tissues due to its sensitivity to the microscopic cellular organization (6). Diffusion based tractography has been performed to identify anatomic connections in mouse brains (7). The derived structural connectivity maps provide insight into the network of interconnected brain regions and ultimately lead to improved diagnosis of various brain disorders (8). Recently, dMRI and tractography in musculoskeletal system has attracted more and more attention to investigate the tissue microstructure, local collagen fiber alignment, and the 3D collagen network (9–11). To the best of our knowledge, noninvasively probing the 3D collagen fiber architecture and connections among different parts of the porcine meniscus have not been reported yet, probably due to a few reasons. First, the tractography is challenging in menisci due to the low fractional anisotropy (FA) and short T2 (4). Second, the spatial resolution is often limited for meniscus MRI due to its thin-layer anatomical structure. Third, the structural connectivity analysis relies on segmenting the meniscus to different sub-regions (12). However, structural connection by diffusion tensor imaging (DTI) reveals not only the local tissue properties but also the 3D fiber network through the whole tissue area.

Besides tractography, the scalar metrics from DTI model are known to be related to the tissue microstructure (8, 13, 14). For instance, the FA and mean diffusivity (MD) are found to be sensitive to the collagen architecture and glycosaminoglycan (GAG) content in different zones of cartilage (15). Exploring the FA and MD variations at different zones of meniscus may help to better understand the microstructure of meniscus, which requires to segment the tissue manually or automatically. Cooper’s classification is one of the most commonly used meniscal classification systems based on the blood supply (16). According to the classification, menisci can be divided into Red-Red zone (outer third of the meniscus), Red-White zone (middle third of the meniscus), and White-White zone (inner third of the meniscus) (17).

Menisci often show low signal intensity on MR images due to the short T2 relaxation time and conventional MRI cannot adequately distinguish red zone and white zone because of little contrast difference between these zones (18). It is possible to visualize enhancement selectively in the red zone of the meniscus using ultrashort echo time (UTE) pulse sequence, but dividing the meniscus to white and red zones is still challenging (19). Conventionally, the meniscus can be manually segmented to different zones, but it is an expertise-intensive and time-consuming process (20). Numerous subjective interpretations for separating adjacent structures with similar image contrasts result in low repeatability and less efficiency. Therefore, automation of the segmentation process is highly desirable.

In this study, we first acquired 3D diffusion-weighed spin-echo pulse sequence to probe the microstructure of porcine meniscus. We then developed a segmentation method with rotational and radial directions to divide the meniscus into different sub-regions. The automatic segmentation method was further validated in the human knee MRI images obtained from the Osteoarthritis Initiative (OAI) database. The water diffusion properties derived from DTI have been quantified at different zones of meniscus. Diffusion tractography was performed through whole meniscus to visualize the 3D collagen fiber network. Combining tractography and automatic segmentation, we were able to observe the structural connections among different areas of the meniscus.

Figure 2 showed the mean volumes and quantitative DTI metrics (FA, MD, AD, and RD) in three different zones. The heterogeneous appearance of the meniscus at different zones was evident in FA image (2A). The volume (2B) gradually increased from W-W zone (10.4%) to R-R zone (53.1%). Similar to the volume, the FA values (2C) gradually increased from W-W zone (0.13) to R-R zone (0.26) with a 100% increase. In contrast, MD, AD, and RD values (2D-2F) gradually decreased from W-W zone to R-R zone. For instance, the RD values changed from 1.0 × 10^-3 mm²/s (W-W zone) to 0.7 × 10^-3 mm²/s (R-R zone), which decreased 30%. Significant differences (p < 0.01) of volume and all the DTI metrics had been found among three different zones. The values of volumes and DTI metrics were summarized in Table 1.

Figure 3 showed the color-FA and the fiber orientation images of meniscus in different regions. The collagen fiber exhibited orthotropic directions between the anterior part (ROI 1, green color) and the body part (ROI 3, red color), between the posterior part (ROI 4, green color) and the body part. The fiber directions were found to gradually change from anterior part to the body part (ROI 2). These fiber directions were also evident from the individual tracts (Figure 4), where the seeding regions (Figure 4A) are from R-R zone (red area), R-W zone (green area), and W-W zone (white area), respectively. The circumferential collagen fibers were found in all three zones.

In order to explore the entire meniscus 3D collagen fiber network, the tractography were performed in the whole meniscus area (Figures 5A, C, front and back view). The tractography showed similar collagen fiber architectures as Figure 4: the circumferential collagen fibers through anterior part to posterior part. However, the whole meniscus tracts revealed the coexistence of both radial and circumferential collagen fibers, especially in ROIs of 1 and 3 (front view), 5 and 6 (back view). The consistent tractography and fiber orientation results were demonstrated in all five menisci (Supplementary Figure 4).

Figure 6 exhibited the automatic segmentation (6A), the diffusion tractography (6B), and the connection heatmap of meniscus (6E). Several distinct characteristics can be identified from the tractography images and heatmap. First, the anterior part showed low connections to the posterior part (red box in 6C), while the body part showed high connections to both anterior part and posterior part (white box in 6C). Second, in the same part (anterior, body, or posterior), the R-W zone exhibited high connections to the adjacent parts (green box and yellow box).

In order to explore the robustness of the automatic segmentation method, the Radial Segmentation was also applied to T2- and T2*-weighted images (Supplementary Figure 5). We further extended our methods for human knee meniscus segmentation (Supplementary Figure 6). Both Radial Segmentation and Rotational Segmentation (Supplementary Figures 6B, C) methods showed visually comparable results to porcine meniscal segmentations.

MRI has emerged as an invaluable component of pathogenesis research in meniscal tear. Quantitative MRI (qMRI) has been applied to study the correlation between the parametric mapping and the severity of tissue degradation (4, 5). Most of the qMRI used to access the biochemical status of the menisci are based on relaxation times, such as T2, T2*, T1rho, and T1, which are challenging to estimate the local collagen fiber alignment directly (24, 25). In order to investigate the comprehensive collagenous fibril texture of meniscus, other imaging modalities such as polarized light imaging (PLM), reflectance confocal microscopy, and scanning electron microscopy (SEM) have been used in previous studies (26–29). While it is well appreciated that complex fiber structures (circumferential and radial) exist within the meniscus using these technologies, they are often limited in a small region and presented as two-dimensional (2D) images. Recently, DTI has been used to study the tissue microstructure and quantify the local collagen fiber direction in knee joint (21, 30–32). The local fiber orientation from DTI model affords an alternative to probe the complex collagen fiber directions of the whole meniscus.

Tractography, as a promising technology to visualize the complex 3D fiber network, has been recently applied to individual connective tissues in knee joint, such as cartilage, anterior cruciate ligament (ACL), and tendon (9, 10, 33, 34). However, the application of tractography to probe the meniscal microstructure is rare, probably due to the low signal-to-noise ratio (SNR) and short T2* values (35–37). To overcome this issue, the menisci were images in a preclinical 7T system with powerful gradients to increase the SNR by minimizing the TE value to 13.0 ms. With the proposed automatic segmentation methods, the connection heatmap has been generated to quantify the connections among different zones (R-R, R-W, and W-W) and different parts (anterior, body, and posterior). Compared to the exisiting qMRI methods focusing on the local change of the tissue properties, the alteration of the connections in meniscus may provide a complementary method to study the integrity of meniscus and meniscal tears in future studies.

It has been reported that FA is sensitive to collagen architecture and MD is sensitive to GAG content in cartilage (31). Cartilage has been found to show depth-dependent response to the degradation, especially in the superficial zone at early osteoarthritis (OA) (18). The variations of DTI metrics at different parts and zones may be related to the meniscal microstructure and composition changes after degradation. Several automatic and semi-automatic segmentation methods have been developed to segment the cartilage, bone, and meniscus in knee joint (38, 39). Although these methods provide excellent segmentation accuracy to distinguish meniscus to other soft tissues, automatically dividing the meniscus to different zones and parts is still limited. Unlike the articular cartilage, which can be divided into three zones according to the collagen fiber directions, both red zone and white zone show similar fiber directions in meniscus (40). The lack of apparent image contrasts between red zone and white zone makes the segmentation more challenging.

In this study, we adapted Cooper’s classification for the segmentation. The segmentation allows us to investigate the diffusion properties of menisci at different zones or parts. It also helps us to obtain the heatmap to show the structural connections among different regions. The proposed segmentation, even is not a gold standard, does afford a convenient way for quantitative analysis of meniscus diffusion properties. The proposed segmentation method requires a simple binary mask of the meniscus, which is relatively easy to obtain from MRI scans, such as T2-weighted image, T2*-weighted image, and DWI (41). The automatic segmentation method may not be limited to MRI and can be applied to different imaging modalities as long as the binary mask is available. Furthermore, the method also shows robust results for human meniscus segmentation, which suggests that our method holds the potential to segment meniscus to different sub-regions for human studies (42).

There are a few limitations in our study. First, the sample size is small due to the extremely long scan time, the consistent collagen fiber alignment and tractography were demonstrated in all five specimens. Second, the dMRI was acquired with a preclinical setting that may not relevant to the clinical study. To the best of our knowledge, it’s still challenging to achieve high quality meniscus dMRI and tractography for in vivo human studies. Advanced acquisition technologies, stronger gradients, higher magnetic field, and novel reconstruction methods shed the light to bridge the gap (43). Third, although the collagen fiber architecture revealed by dMRI is consistent with other imaging modalities in previous studies, complete validation of MRI findings using other imaging modalities is warranted in future studies. Last, this Cooper’s classification is originally defined for human meniscus, our method may be improved for animal meniscal studies when further shape information about different species is known.

In conclusion, the porcine meniscus microstructure was investigated using a 3D diffusion-weighed spin-echo pulse sequence. Strong zonal-dependent diffusion properties were demonstrated by DTI metrics (FA, MD, AD, and RD). The complex 3D collagen fiber architecture of the entire meniscus was visualized by diffusion tractography. Combining tractography and automatic segmentation method, we were able to observe the structural connections among different areas of the meniscus. It may offer a novel method to evaluate the local meniscus tears and address the alteration of connections among different regions of the meniscus.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The animal study was reviewed and approved by institutional animal care and use committee (IACUC).

JS: Methodology, Formal analysis, Writing-review and editing. QZ: Editing, Formal analysis, Validation. YQ: Specimen preparation, Methodology, Writing-review and editing. GC: MRI acquisition, Writing-review and editing. GJ: Conceptualization, Investigation, Writing-review and editing, Supervision, Funding acquisition. NW: Conceptualization, Methodology, MRI acquisition, Formal analysis, Investigation, Writing -original draft, Funding acquisition, Writing-review and editing. All authors contributed to the article and approved the submitted version.

This work was supported by the NIH/NIBIB National Biomedical Technology Resource Center P41 EB015897 (to GJ), Charles E. Putman MD Vision Award of the Department of Radiology, Duke University School of Medicine (to NW and Charles E. SPritzer), and Strategic Research Initiative (SRI) IUH and Indiana University School of Medicine (to NW).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.