Predict Axillary Node Metastasis or Disease-Free Survival with MRI Radiomics

December 9, 2020
Whitney J. Palmer

The tools incorporate a patient’s own clinical characteristics to identify which individuals with early-stage breast cancer will experience cancer spread and which will potentially have disease-free survival.

MRI radiomics and a patient’s own clinical risk characteristics can be used to predict which axillary lymph nodes will metastasize and which patients with early-stage breast cancer will experience a disease-free survival, new research has revealed.

In a study published on Dec. 8 in JAMA Network Open, a group of investigators from China outlined two clinical-radiomic nomograms that are highly accurate in predicting metastasis and categorizing patients by their level of risk.

Currently, said the team led by Yunfang Yu, M.D., from Sun Yat-sen Memorial Hospital in Guangzhou, providers use sentinel lymph node biopsy to determine whether an axillary lymph node will metastasize. Knowing whether cancer has spread from a node plays a critical role in evaluating a patient’s prognosis, but this biopsy method is both invasive and can lead to a high false-negative rate.

Related Content: Radiomics Could Be the Next Crime Scene Investigator

But, to date, no non-invasive technique for analyzing and predicting axillary lymph node metastasis or potential disease-free survival has existed. To fill that void, Yu’s team ventured to create and validate dynamic contrast-enhanced MRI radiomic features.

“The clinical-radiomic nomograms were useful in clinical decision-making associated with personalized selection of surgical intervention and therapeutic regiments for patients with early-stage breast cancer,” the team said.

To develop these clinical-radiomic nomograms, Yu’s team retrospectively collected data from 1,214 women who were treated for histologically confirmed early-stage breast cancer in one of four hospitals in China between July 3, 2007, and Sept. 21, 2019. All patients underwent pre-operative MRI and were treated with surgery and sentinel lymph node biopsy or axillary lymph node dissection, and for the study, the team divided the women between a development group (70 percent) and a validation group (30 percent).

Using machine learning algorithms that can evaluate extracted radiomics features, the team created an MRI radiomics signature for the prediction of axillary lymph node metastasis, as well as disease-free survival. Based on their calculations, to identify node metastasis, the signature produced an area under the curve (AUC) of 0.88 and 0.85 for the development and validation groups, respectively. In addition, it predicted three-year disease-free survival with an AUC of 0.81 for the development group and 0.73 for the validation group.

Related Content: Radiomics Gathered from CT Scans Predict Outcomes from COVID-19 Pneumonia

Alongside the MRI radiomics signature, Yu’s team also created a clinical signature based on two clinical-radiomics nomograms – one for metastasis and one for disease-free survival – that brought clinical risk factors into the fold. Overall, the team said, the decision curve analysis showed that the clinical-radiomic nomogram had more clinical predictive usefulness than did either the clinical or radiomic signatures by themselves.

But, these tools are not yet ready for clinical prime time, said Nathaniel Braman, Ph.D., a biomedical engineer with Case Western Reserve University. More work still needs to be done to confirm how they perform with data that was not used during training, he said in an accompanying editorial. Further research should include information gleaned from different scanners, using different acquisition protocols, and performance should be assessed using international populations, as well.

To reach the clinical implementation goal, he said, the team will need to replace manual image annotation with a fully-automated segmentation approach that can allow for greater generalization and high performance or employ an in-depth analysis of the impact of variability between readers on its performance. Such a model will need to be selected and evaluated on new data, he added.

It is also vital, he said, that the signature be prospectively validated for the guiding surgical approach, as well as compared with sentinel lymph node biopsy for any improved outcomes.

“The retrospective performance of radiomic decision support tools will ultimately be immaterial unless they can also yield tangible improvements to quality of care for patients whose treatment has reached a crossroads,” he concluded.

For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-newsletter here.