Effect of Bone Reading CT software on radiologist performance in detecting bone metastases from breast cancer.

Abstract

To evaluate the effect of CT software that generates rib unfolding images and automatically numbers ribs and thoracic spines on radiologist performance in detecting thoracic bone metastases from breast cancer. A total of 126 patients with breast cancer who underwent chest CT and fludeoxyglucose (FDG)-positron emission tomography (PET)/CT and/or bone scans were retrospectively reviewed. One board-certified radiologist (R1) and one radiology resident (R2) independently assessed the original chest CT and rib unfolding images using a commercially available post-processing software (Bone Reading) application to evaluate metastasis in the ribs and thoracic spines. Results were compared with reference standard based on CT, FDG-PET/CT and/or bone scan with follow-up. Based on reference standard, 78 metastatic bone lesions in 26 patients were identified. On per-patient-based analysis, Bone Reading assessed by R1/R2 had a sensitivity of 84.6%/80.8% and a specificity of 94.0%/94.0% with an accuracy of 92.1%/91.3%. The original CT reading yielded a sensitivity of 73.1%/57.7% and a specificity of 95.0%/94.0% with an accuracy of 90.5%/86.5%. The sensitivity and accuracy of Bone Reading were significantly higher than those of CT reading, as assessed by R2 (both p = 0.031). On per-lesion-based analysis, Bone Reading assessed by R1/R2 yielded a sensitivity of 84.6%/82.1% and a specificity of 99.7%/99.6% with an accuracy of 99.4%/99.3%, while the original CT reading yielded a sensitivity of 71.8%/62.8% and a specificity of 99.6%/99.5% with an accuracy of 99.2%/98.9%. The sensitivity and accuracy with Bone Reading application were significantly higher than those with CT reading by both readers (R1, p = 0.006 and p = 0.036, respectively; R2, both p < 0.001). The mean reading time needed for Bone Reading application was significantly shorter than that for original chest CT reading (p < 0.001). Bone Reading application helped readers find small and sclerotic lesions missed in original CT reading. In patients with breast cancer, the use of Bone Reading application improved radiologist performance in bone metastasis detection compared with original chest CT reading with reduced reading time. This software will be more helpful to inexperienced radiologists for improving the reading performance. Advances in knowledge: Small and sclerotic lesions can be easily missed in original CT reading. Using Bone Reading CT software can enhance the performance of radiologists in detecting bone metastasis in breast cancer. False-negative rates can be significantly reduced in both inexperienced and experienced readers.