Is Strain Elastography Really a Good Adjunct for Prediction of Malignancy in Soft Tissue Tumours?

Dear Editor,

With great interest we have studied the recent publication by Riishede et al. (Riishede et al. Ultraschall in Med 2015; 36: 369 – 374) concerning the role of sonoelastography (SE) as an adjunct diagnostic mean in the differentiation of soft tissue tumours. SE is certainly a new method on the rise as can be concluded from the increasing number of publications. However, its clinical adaptation seems to lag behind scientific propositions and is still mainly confined to hepatic, breast, thyroid and prostate applications (Dudea S. Med Ultrason 2014; 16: 87 – 88). To our opinion not without reason, with strong operator-dependency and sometimes erratic inter-observer agreement (Carlsen JF et al. Diagnostics 2013; 3: 117 – 125). Regardless of the slow acceptance in clinical routine, numerous papers continue to be published examining the use of SE in various other settings, even in the assessment of intrapulmonary nodules (Adamietz B et al. Ultraschall in Med 2013; 35: 33 – 37). One of such is the recent article by Riishede et al. which proposes SE as a viable addition in the diagnosis of soft tissue tumours. However, we believe that the proposition is poorly substantiated and that the actual information presented should rather lead to an opposite conclusion.

Strain SE does not yield absolute values for elasticity such as SI-based Pascal, but rather returns relative compressibility given as false-colour overlays. Thus strain ratios between regions of interest (ROIs) are usually calculated to compare target tissue with surrounding reference tissue – preferably at the same depth as the primary ROI (Carlsen JF et al. Diagnostics 2013; 3: 117 – 125). Obviously the tissue elasticity within the reference ROI influences the strain ratio and should be chosen with care. In contrast to rather homogeneous organs such as the liver or breast, the surrounding tissue in musculoskeletal lesions varies widely and can range from bone, muscles or tendons to subcutaneous fat. This is not properly accounted for in the study and not mentioned in the discussion.

The figures and quantitative analyses presented in the paper show several inconsistencies. Visual scoring – which is commonly used to grade tissue elasticity (Pedersen M et al. Ultraschall in Med 2012; 33: 441 – 446) – was demonstrated to be non-diagnostic. The ratio of malignant to benign soft tissue tumours was roughly the same across SE groups (modified from Tsukuba (Itoh et al. Radiology 2006; 239: 341 – 350)) with about 40 %; only the second group scored a bit higher with 56 %, but due to the relatively small number of cases this is not surprising (data from figure 6c: resp. 3 vs. 8, 9 vs. 16, 4 vs. 10, 3 vs. 8). Curiously the (ancient) schwannoma in Fig. 3 is presented as an exemplary benign tumour, while the deeper cystic-like area is deeply blue in the elastogram. In our view this does not make much sense, since schwannomas can often be identified due to central cystic areas (as is the case here) and neural contact. All in all, this undermines one of the authors’ main statements that visual grading does not allow for tumour classification and – from a practical standpoint – only serves to demonstrate that SE carries no information beyond the obvious in this case.

Strain ratio on the other hand apparently demonstrated a significant difference between the groups; the authors report p = 0.043. However, totally contradictory to this result, the presented confidence intervals of the mean strain ratios of the two groups would lead us to the conclusion that there is no statistical significance at a two-sided < 0.05 level (1.94 [95 % CI 0.37 to 10.21] and 1.35 [95 % CI 0.32 to 5.63]). The p-value will be less than 0.05 only when the 95 % confidence interval does not contain the other mean strain ratio (see § 8.8.1 in: "Practical statistics for medical research" (Altman DG. Chapman and Hall, 1991)). This condition is not fulfilled here. In addition, the comparison between box plots for benign and malignant tumours (all 61) in figure 6a does not give reason to believe there is a relevant difference.

The authors furthermore report an increase in difference of mean strain ratios for benign and malignant lesions when excluding fat-containing tumours. As fat-containing lesions are the largest group of musculoskeletal tumours, usually present as tender masses and frequently prove difficult to diagnose as certainly benign or malignant, the motivation behind their exclusion is not clear. On the other hand, cases of osteosarcoma and chondrosarcoma were included, which are typical ‘hard’ lesions containing osteoid or chondroid matrix and are often readily diagnosed using CT and MRI.

Overall there is no clear support for the purported usefulness of SE in soft tissue tumours as the data presented by Riishede et al. do not demonstrate significance for either visual grading, histogram analysis or strain ratios. Shear-wave SE may prove better suited for such an application, but until further evidence-based research is presented there is no incentive to use SE in the diagnosis of soft tissue tumours.

• References

• 1 Riishede I, Ewertsen C, Carlsen J et al. Strain Elastography for Prediction of Malignancy in Soft Tissue Tumours – Preliminary Results. Ultraschall in Med 2015; 36: 369-374
• 2 Carlsen J, Ewertsen C, Sletting S et al. Ultrasound Elastography in Breast Cancer Diagnosis. Ultraschall in Med 2015;
• 3 Cosgrove D, Piscaglia F, Bamber J et al. EFSUMB Guidelines and Recommendations on the Clinical Use of Ultrasound Elastography. Part 2: Clinical Applications. Ultraschall in Med 2013; 34: 238-253 DOI: 10.1055/s-0033-1335375. Accessed October 11, 2013
• 4 Itoh A, Ueno E, Tohno E et al. Breast Disease: Clinical Application of US Elastography for Diagnosis1. Radiology 2006; 239: 341-350 http://radiology.rsna.org/content/239/2/341 Accessed July 19, 2012
• 5 Krzywinski M, Altman N. Points of significance: error bars. Nat Methods 2013; 10: 921-922