AI predicts how breast tumors will respond to chemotherapy

Artificial intelligence (AI) can help predict how a breast tumor will respond to neoadjuvant chemotherapy (NAC), according to new findings published in the Journal of Digital Imaging.

“Advances in genomics have demonstrated breast cancer to be a disease with a spectrum of biologically relevant molecular subtypes,” wrote Richard Ha, MD, department of radiology at Columbia University Irving Medical Center in New York City, and colleagues. “This significant disease heterogeneity poses a major challenge in the development of novel treatments. Targeted therapies may only be effective in a small subset of breast cancers, which has contributed to the difficulty establishing a therapeutic benefit in a large, heterogeneous clinical trial.”

The authors used a breast MRI tumor dataset to train a convolutional neural network (CNN). A total of 141 breast cancer patients from January 2009 to June 2016 were selected. All patients underwent breast MRI prior to NAC, completed Adriamycin/taxane-based NAC and underwent surgical resection. These patients were divided into three groups based on how the tumor responded to NAC: complete (group 1), partial (group 2) and no response (group 3).

The CNN’s overall accuracy was 88 percent. Breaking the performance down by groups, group 1 had a specificity of 95.1 percent  ± 3.1 percent, sensitivity of 73.9 percent  ± 4.5 percent and accuracy of 87.7 percent ± 0.6 percent. Group 2 had a specificity of 91.6 percent  ± 1.3 percent, sensitivity of 82.4 percent  ± 2.7 percent and accuracy of 87.7 percent ± 0.6 percent. Group 3 had a specificity of 93.4 percent  ± 2.9 percent, sensitivity of 76.8 percent  ± 5.7 percent and accuracy of 87.8 percent ± 0.6 percent.

“Our results demonstrate that it is feasible to utilize CNN to predict NAC response prior to initiation of therapy,” Ha and colleagues wrote. “This represents an improved approach to early treatment response assessment based on a baseline breast MRI obtained prior to the initiation of treatment and significantly improves on current prediction methods that rely on interval imaging after the initiation of therapy.”

The authors did note that their research was based on a small number of cases from a single institution, adding that a larger dataset “will likely improve our prediction model.”