A Phantom Study for Assessing the Effect of Different Digital Detectors on Mammographic Texture Features

Digital mammography (DM) is commonly used as the breast imaging screening modality. For research based on DM datasets with various sources of x-ray detectors, it is important to evaluate if different detectors could introduce inherent differences in the images analyzed. To determine the extent of such effects, we performed a study to compare the effects of two DM detectors, the GE 2000D and DS, on texture analysis using a validated breast texture phantom (Yaffe et al., University of Toronto). DM images are acquired in Cranio-Caudal (CC) view, and texture features are generated for both raw and post-processed DM images. Image intensity profiles and texture features are compared between the two detector systems. Our results suggest that there are inherent differences in the images. For raw and processed images, the image intensity cumulative distribution function (CDF) curves reveal that there is a scaling and shifting factor respectively between the two detectors. Image normalization with z-score can reduce detector differences for grey-level intensity and the histogram-based texture features. The differences between co-occurrence and run-length texture features persist after intensity normalization, suggesting that simple z-scoring cannot alleviate all the detector effects, potentially also due to differences in the spatial distribution of the intensity values between the two detectors.