Radiomic Analysis of Sonograms for the Detection of Tendon Damage

Abstract

The exact precursors to tendinopathy can be multiple and nuanced. Further complicating effective diagnosis is the reliance on tendon pain as a measure of injury. Even severe tendinopathies can remain asymptomatic as overuse damage accumulates and diagnosis using quantitative indicators of damage is challenging. Hence, there is a critical need to establish biomarkers to assist in early detection of tendinopathy in at-risk populations such as athletes. One set of imaging biomarkers can be generated using radiomics. This technique extracts numerical values from medical images, such as those captured using ultrasound (US), for clinical decision-making. US images are the standard for confirming clinical tendinopathy diagnoses, as they are more accurate, less time-consuming, and less expensive than other modalities. Texture analysis, a subset of radiomics, uses algorithms to quantify the shade, alignment, and distribution of pixels within US images to assess the disease state of a tissue. Prior clinical studies of texture analysis have shown that it can reliably identify differences in tissue structure and composition. In symptomatic cases of tendinopathy, tendon US images appear darker and more disordered. One drawback to previous in vivo studies, however, is the large variation of higher order statistics between US images, likely due to changes in ultrasound transducer operation or patient positioning. To circumvent these limitations, imaging of an ex vivo tendon will be conducted simultaneously with loading in this study. A total of 14 juvenile, female porcine superficial digital flexor tendons were excised and loaded using an MTS Insight 10 load frame. US imaging was carried out using a Supersonic Imagine Aixplorer SLH20-6 transducer. After preloading at 2 N, each sample in the progressive strain group (n = 6) was subjected to progressively increasing strain levels (2, 4, 6, and 8%), utilizing a 25-minute stress relaxation at each strain magnitude with 20-minute recoveries. Once the stress stabilized, the equilibrium stiffness (Eeq) was calculated, and additional US images were taken. The onset of tendon damage was defined by a reduction in Eeq, as well as other mechanical parameters such as laxity and percent relaxation. The strain level at which this damage occurs informed extension parameters for the second group (n = 8), where US images were acquired before and after applying a single 10% strain ramp to induce damage. It was found that numerous parameters describing image texture correlated with reductions in tendon mechanical properties, with higher order parameters generally correlating more strongly. This novel data suggests that higher order parameters can serve as reliable imaging biomarkers of matrix damage and can be used in the future for clinical diagnostics.