Objective: The aim of this study was to determine if the presence of focal normal bone marrow fat signal within a tumor on magnetic resonance imaging excludes malignancy.
Materials and methods: One hundred eighty-four histologically proven tumors with available magnetic resonance imaging (MRI) of the appendicular skeleton and pelvis from 184 patients were collected and reviewed at two separate institutions. There were 111 malignant and 73 benign tumors. Two radiologists at each institution, blinded to the diagnosis, reviewed the MRIs independently and reported the presence or absence of normal marrow fat signal within the tumor based upon T1-weighted imaging without fat suppression and T2-weighted imaging with fat suppression and/or short inversion-time inversion recovery (STIR). Discrepancies were then reviewed in consensus to determine the presence or absence of focal normal marrow signal. For each institution, a Fisher's exact test was used to compare the frequency of focal normal marrow fat signal in benign and malignant tumors. This comparison was performed for each reader, as well as for the consensus reading at each site. Positive and negative predictive values were also calculated for each reader, as well as the consensus reading at each site. Fisher's exact test was also used to compare the frequency of intratumoral fat in benign and malignant lesions for the pooled sample. Bayes theorem was used to calculate the positive and negative predictive values for the pooled consensus data. Ninety-five percent confidence intervals were constructed for the pooled estimates using a bootstrapping algorithm.
Results: There was good interobserver reliability of 95.3% and 96.7% at sites 1 and 2, respectively. There were three discrepancies (one malignant and two benign) at site 1 and four discrepancies (two malignant and two benign) at site 2. Reader consensus at site 1 identified normal marrow fat signal within 1 of 50 (2.0%) malignant and three of 14 (21.4%) benign tumors. Findings were statistically significant with a p value of 0.030. The positive predictive value (PPV) and negative predictive value (NPV) at site 1 was 81.7% and 75.0%, respectively. Reader consensus at site 2 identified normal marrow fat signal within three of 61 (4.9%) malignant and 14 of 59 (23.7%) benign tumors. Findings were statistically significant with a calculated p value of 0.004. The PPV and NPV at site 1 was 56.3% and 82.4%, respectively. For the pooled consensus, the frequency of intratumoral fat in benign lesions (17/73, 23.3%) is significantly greater than the frequency in malignant lesions (4/111, 3.6%), p < 0.001.
Conclusion: The presence of focal normal marrow signal within a tumor is highly suggestive of a benign tumor.