ABSTRACT
OBJECTIVE: To evaluate the diagnostic performance of unenhanced MRI (UE-MRI) for malignant breast lesions and its reproducibility. PATIENTS AND METHODS: We retrospectively included 118 patients who had breast MRI. DWI and STIR images were read in combination and referred to as UE-MRI; the presence or absence of the malignant lesion was noted by two observers. Their results were compared with those of final histopathology or with a two-year negative follow-up for diagnostic performance assessment; ROC curves were built. Diagnostic performance was stratified according to lesion site and size. Interobserver agreement was evaluated through the Cohen's k statistic. RESULTS: Specificity of STIR and DWI was 99.3% and 95.7% for Reader 1; 99.3% and 96.4% for Reader 2. Sensitivity was 76.5% and 76.5% for Reader 1; 77.5% and 77.6% for Reader 2. The ROC AUC (Reader 1) was 0.869 and 0.844 for STIR and DWI, respectively (p<0.001 both); for Reader 2, values were 0.874 and 0.853 respectively (p<0.001 both). Lesion dimension ≤10 mm was associated with lower AUC values. Lesion site didn't influence the diagnostic performance. Interobserver agreement was very good for STIR and DWI (k=0.887, p <0.001, and k=0.867, p <0.001). DISCUSSION: UE-MRI has a good overall diagnostic performance in the detection of breast cancer and a very good specificity for both STIR and DWI sequences. We observed reduced diagnostic performance for lesions ≤10 mm in size. Lesion's site isn't associated with a significantly decreased diagnostic performance of UE-MRI. There's a good interobserver agreement for both sequences (STIR and DWI). CONCLUSIONS: UE-MRI may be employed in patients with contraindication to gadolinium. It has considerable specificity and positive predictive value and good reproducibility.
