Breast cancer is the leading cause of cancer death among women, commonly due to the presence of distant liver metastases. CT and MRI can be used to evaluate treatment response based on anatomic and morphologic changes, but measuring lesions dimensions can be subject to inter- and intraobserver variability, according to presenter Dr. Leila Mostafavi of Massachusetts General Hospital in Boston. This can lead to inconsistent assessments.
Seeking to determine if radiomics could help assess and predict treatment response, the researchers retrospectively identified 203 adult women with hepatic metastases from breast cancer who were treated with chemotherapy. Each patient had two baseline CT exams, with follow-up performed on seven different CT scanners from four vendors. Next, they applied a machine learning-based semiautomatic segmentation and radiomics software prototype (Siemens Healthineers) to the CT studies.
Despite substantial variations in the scanners, acquisition, and reconstruction parameters used for these patients, radiomics showed a high accuracy -- an area under the curve (AUC) up to 0.89 -- for differentiating stable hepatic metastases from partial response or progressive disease on baseline and follow-up CT scans, according to the researchers. However, radiomics could not predict disease progression or partial response based on baseline images alone, only producing an AUC up to 0.58.
"The key implication of our study is the feasibility of semiautomatic lesion segmentation and accuracy of radiomics, which has the potential to replace or supplement conventional manual lesion measurement and lesion dimensions for assessing treatment response," she told AuntMinnie.com. "Although minor adjustment in lesion contours was needed in most cases, the editing effort took less than a minute per exam."
Check out this presentation on Sunday to learn more.
