Accuracy of magnetic resonance imaging in planning the osseous resection margins of bony tumours in the proximal femur: based on coronal T1-weighted versus STIR images.

Abstract

Assessment of the extent of tumours using magnetic resonance imaging (MRI) is the basis for bone resection in limb-salvage surgery. We aimed to compare the accuracy of T1-weighted MRI and STIR sequences in measuring the extent of proximal femoral tumours, using the macroscopic specimens as the gold standard for comparison. We compared single coronal T1-weighted with STIR sequences in 34 proximal femoral tumours, using bivalved resected macroscopic tumours for comparison. After randomisation, four observers measured longitudinal osseous tumour extent using MRI and specimen photographs on two separate occasions, 3 weeks apart. There were 25 metastatic tumours, 8 chondrosarcomas and 1 myeloma. Eight patients presented with pathological fractures. The Pearson's correlation coefficient for comparison of T1 with macroscopic tumours was 0.91 (95% confidence interval [CI]: 0.83 to 0.96) for all observers and 0.90 (95% CI: 0.81 to 0.95) for STIR images. This difference was not statistically significant, and T1 and STIR sequence measurements had similar precision and accuracy. Bland-Altman plots showed T1-weighted imaging to be unbiased, whereas STIR sequences were biased and had systematic error. Moreover, STIR measurements overestimated tumour size by 6.4 mm (95% CI: -26.9 to 39.7 mm) and 2 patients were outliers. T1 measurements were closer to the macroscopic measurements with a mean difference of 1.3 mm (95% CI: -28.9 mm to 31.5 mm), with 3 patients falling outside of this. The variance was greater for STIR measurements. This difference between T1 and STIR measurements was statistically significant (p = 0.000003). The intra-observer reliability between separate measurements for MRI and specimen photographs achieved interclass correlation coefficients of 0.97, 0.96 and 0.95 (T1, STIR and macroscopic tumour respectively). T1 had greater interobserver correlation than for STIR and macroscopic tumour measurements (0.88 vs 0.85 and 0.85 respectively). These differences in interclass correlation were not statistically significant. This study has shown T1-weighted MRI sequences to be unbiased compared with STIR sequences at determining intra-osseous tumour extent. STIR overestimates the length of bone tumours. T1 is therefore preferred for pre-operative planning for the resection of bone tumours.