This proof-of-concept exploratory study used retrospective MADI data from samples previously acquired by the MADI research group from Imperial College London (Figure 1) [4]. Four human ACL samples were analysed, representing a range of tissue conditions: healthy, partially torn, completely torn and reconstructed ligaments. All data was anonymised prior to analysis. Ethical approval for secondary analysis was confirmed under the framework of the original study by Imperial College London Ethics Committee.

Segmentation and Vector Field Generation:

Regions of interest corresponding to the ACL were manually segmented using 3D Slicer (v5.8.0)[12]. Segmentation was performed across multiplanar views using greyscale volumes generated by averaging voxel intensity across all acquisition orientations. This ensured collagen exhibiting the magic angle effect at different orientations was captured. Volumes were acquired using a prototype low resolution MADI scanner with strength 1.5T[4,5]. Spatial resolution of captured volumes was 1mm³. Ambiguous tissue boundaries were segmented following known anatomical constraints. Existing segmentation masks, developed by the MADI research team, were reviewed and refined where necessary in line with an agreed segmentation protocol.

Collagen orientation vectors for each 1mm³ voxel were calculated using the established MADI processing pipeline[4]. Each voxel was represented by a unit vector confined to a single hemisphere to remove directional ambiguity of collagen fibres.

Metric Development

FC was developed to quantify local collagen alignment between neighbouring voxels within the AC (Figure 2)L. The FC metric was developed using MATLAB (vR2024b) from MADI generated vector fields[13]. It was then validated using in silico vector fields with known ground truth generated in MATLAB to simulate aligned and disrupted collagen fibres under controlled conditions (Figure 3)[14]. This was performed over multiple noise conditions to mimic orientation uncertainty introduced by measurement noise and reduced signal-to-noise ratio, by applying controlled angular perturbations to vector orientations. For each target voxel, FC was calculated by comparing its orientation vector with surrounding vectors in a 3x3x3 voxel cubic neighbourhood. The final FC score for the target voxel was then calculated by computing the angular difference in degrees between each the mean neighbourhood orientation vector and the target voxel’s orientation vector. Therefore, the FC value was represented with a range between 0° to 90° with higher FC values indicating greater deviation from local uniform alignment, reflecting potential disruption to native collagen structure associated with injury.

Biological Application and Analysis

After validation FC was applied to the MADI-derived vector fields representing the 4 sample human ACLs. Quantitative assessment of local tissue organisation included analysis of spatial autocorrelation using Moran’s I to evaluate global FC patterns[15]. Qualitative assessment was conducted with the aid of an MSK consultant radiologist with 13 years of experience, against expected patterns from the various tissue conditions.

Visualisation

Voxel-wise FC values were rendered as 3D scalar colour maps and overlaid onto streamline tractography generated in 3D Slicer. This produced hybridised visualisations integrating quantitative FC information with fibre pathway representations.