| This study compared different cumulative maximum intensity (CMI) image processing techniques for depicting vascularity in subharmonic ultrasound images (SHI) of breast lesions. In CMI mode a composite image depicting vascular architecture and blood flow is constructed through maximum intensity projection accumulation of SHI data over consecutive images. SHI data (transmitting and receiving at 4.4/2.2 MHz) was obtained from 16 breast lesions using a modified Logiq 9 Scanner (GE Healthcare, Milwaukee, WI). CMI data was processed manually and by an automated technique using 3 different threshold values to reduce motion artifacts. In every case an image of peak flow was also selected as control. Six blinded and independent readers scored all the randomized images for vessel continuity, detail resolution, presence of artifacts, overall image quality and SNR on a 7 point scale (poor-excellent). Readers subsequently ranked all images within each case on a scale from 1-5 (best-worst). Scores were compared using a double, repeated measures ANOVA. The processing techniques were significantly different with regards to vessel continuity, detail resolution and image quality (p<0.001). The single (control) frame was scored significantly worse compared to the manual and automated techniques (p<0.001). For all parameters assessed significant differences were observed between users (p<0.001). One of the users differed markedly from the rest, but repeating the analysis without this user yielded similar results. The rank obtained by each of the techniques (control: 3.98 ± 1.38; manual CMI: 3.45 ± 1.31; automatic CMI: 2.70 ± 1.46 to 2.97 ± 1.35; averaged over all users with lower scores being better) was also significantly different (p<0.025). The automated CMI techniques processed data faster and were more reproducible than the manual method. In conclusion, CMI processing techniques visualize vascularity and produce better image quality than the best single SHI frame. Moreover the automated techniques save processing time, eliminate user bias and are more reproducible than the manual method. |