Researchers at the Seattle Cancer Care Alliance and University of Washington Medical Center (Seattle, WA) have used the CADstream automated MRI analysis system from CAD specialist Confirma (Bellevue, WA) to retrospectively examine MRI scans of suspicious breast lesions. The software is designed to automate the manual image processing and analysis currently performed by technologists and radiologists.
The research team examined MRI scans of 154 breast lesions (41 malignant, 113 benign) from 125 women. The lesions, which were identified and biopsied between 2001 and 2004, were only visible using MRI (nonpalpable and not seen in a mammogram). The scans were processed using CADstream version 3.0 and the results compared with the original findings and recommendations made by radiologists.
"There are challenges associated with breast MRI and one is the time it takes to process and evaluate the many images acquired," said Constance Lehman, director of radiology at the Seattle Cancer Care Alliance. "Computer software programs such as the one evaluated in our study can assist us in interpreting breast MRI scans more easily."
Data deluge
MRI is used as a mammography adjunct, to provide an enhanced view of suspected malignant tissue, examine the contralateral breast in patients with newly diagnosed breast cancer and screen women at high risk for breast cancer. Typically, the imaging process includes one MRI scan without contrast agent and two more after the contrast is administered.
One key challenge when interpreting breast MR images is the assessment of a lesion's morphology and kinetic features (the amount of contrast agent absorbed by the tissue over time) on these multiple imaging series. To address this issue, the researchers used CADstream to automatically assess the kinetic features for each lesion.
The software incorporates the three series of MRI data - pre-contrast, immediate post-contrast and delayed post-contrast - in its calculations. It first compares the pixel signal-intensity values on the pre-contrast and immediate post-contrast MRI scans. If a pixel's value increases above a user-specified threshold, in this case 50% or 100%, it is identified as "meeting threshold enhancement".
For such pixels, the program then compares signal-intensity values on the immediate and the delayed post-contrast series. Each pixel is colour-coded according to whether its value decreases by more than 10% on the delayed series (red), increases by more than 10% (blue) or changes by less than 10% (green). The resulting colour map overlaid on the MR image indicates the regions of threshold enhancement.
Analysis of the tissue enhancement data as displayed in the CAD-processed images identified 38 of the 41 malignant lesions, using both the 50% and 100% enhancement thresholds. At the 50% threshold, the number of false positives was reduced by 8.8%, compared with those at the original interpretation. When the 100% threshold was applied, false positives went down by 23%.
The authors concluded that "computer-aided evaluation has the potential to improve the discrimination of benign from malignant breast lesions at MR imaging". They also noted that, while computer-aided evaluation is useful as a tool to supplement the radiologist's subjective interpretation, it should not be relied upon exclusively to guide patient management.