Development and evaluation of a soft-copy mammographic viewing protocol to improve radiological reporting

[摘要] English: INTRODCUTION: Switching from screen-film mammography to digital mammography entails a lot more for the reporting radiologists, than switching from a light box to a computer monitor. Soft-copy viewing of the digitally processed image demands different skills and thus knowledge from the radiologist. The image processing option on digital mammography units is vendor dependant and the optimal processing options have not yet been established. The main aim of this study was to develop and evaluate a soft-copy viewing protocol for mammography through participative learning to improve radiological reporting. METHODS A phantom-based method was used to identify a smaller set of processing options to be evaluated for image quality assessment on clinical images. Three (3) radiologists were trained in the new modality with specific emphasis on how to address the challenges of soft-copy viewing. The viewing protocol was developed through participative learning. The radiologists scored the image quality on thirty six (36) medio-lateral oblique images processed with four (4) different image processing options (MUSICA2, MUSICA2 Invert, Unprocessed, and Unprocessed Invert). An image quality score was calculated to find the best processing option for the anatomical structures overall, anatomical structures individually, masses, calcifications, noise, and the early detection of breast cancer. A viewing protocol was recommended based on the findings. The effect of the viewing protocol was assessed by comparing diagnostic accuracy of the radiologists before and after the viewing protocol. They reported on eighty (80) mammograms using the breast imaging and reporting data system (BI-RADS) of the American College of Radiology. Sensitivity, specificity, positive predictive value (PPV) and BI-RADS category 3 were calculated and compared. RESULTS The phantom-based method found Unprocessed Invert, MUSICA2, MUSICA2 Invert, and Unprocessed to provide the best image quality. These processing options were therefore identified for image quality assessment on clinical images. For the anatomical structures overall, MUSICA2 provided significantly superior image quality compared to Unprocessed (p<0.0001) and Unprocessed Invert (p<0.0001). MUSICA2 Invert also provided significantly superior image quality compared to Unprocessed (p<0.0001) and Unprocessed Invert (p=0.0003) for that. The only significant difference between MUSICA2 and MUSICA2 Invert was found for skin outline for which MUSICA2 Invert showed superiority (p=0.0563). The image quality of vessels in dense parenchyma was found be significantly inferior to that of all other anatomical structures with all processing options, even with the processed images (p<0.0001). For calcifications MUSICA2 provided significantly superior image quality compared to Unprocessed and its Invert (p=0.0066 and p=0.0001 respectively). However, no significant difference was found between any of the processing options for masses (p>0.05). Noise was significantly less visible for Unprocessed compared to MUSICA2 (p = 0.016) although it was still acceptable to all three radiologists in 97.2% of cases with MUSICA2. For the early detection of breast cancer, MUSICA2 was found to be significantly superior to Unprocessed (p=0.0003) and Unprocessed Invert (p=0.0005). The recommended default processing option for the viewing protocol was MUSICA2 Invert. After the development of the viewing protocol, sensitivity increased for two of the radiologists [from 90% to 95% (p=0.6752)], and from 90% to 97.5% (p =0.3589) respectively]; specificity increased for two of the radiologists [from 61.5% to 72.5% (p=0.2999), and from 70% to 85% (p=0.1082) respectively]; PPV increased for all three radiologists [from 71.7% to 77.6% (p=0.6198), from 75% to 86.4% (p=0.1699), and from 83.7% to 84.8% (p=0.8907) respectively]. The percentage BI-RADS category 3 cases decreased for two of the radiologists [from 15% to 12.5% (p=0.6461) and from 28.8% to 22.5% (p=0.2810) respectively]. CONCLUSIONS Although not significant, the study found improvement in diagnostic accuracy after the development of the viewing protocol. Training of radiologists in the new modality and knowledge of the effect of image processing on image quality is regarded as important. The development of the viewing protocol through participative learning of the radiologist provided evidence to the radiologists that they could confidently use the proposed viewing protocol in clinical practice.

[发布机构] University of the Free State