MRI-trained algorithm can predict breast tumor response to chemotherapy

Using a breast MRI tumor dataset, researchers found a deep learning convolutional neural network (CNN) approach could be trained to predict response to chemotherapy prior to its initiation, according to a recent Journal of Digital Imaging study.

“Our results demonstrate that it is feasible to utilize CNN to predict neoadjuvant chemotherapy (NAC) response prior to initiation of therapy,” wrote lead author Richard Ha and colleagues. “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.”

Ha, with Columbia University Irving Medical Center in New York, and colleagues identified 141 locally advanced breast cancer patients in their database from January 2009 to June 2016. All had undergone breast MRI before NAC, successfully completed Adriamycin/ taxane-based NAC and underwent surgical resection with final surgical pathology data available. Patients were divided into three groups based on their NAC response: complete (group 1), partial (group 2) and no response/ progression (group 3).

The team evaluated 3,107 volumetric slices of 141 tumors. Overall, the CNN achieved 88 percent accuracy over the three classes of NAC response. Additional results were as follows:

  • Group 1 achieved a specificity of 95 percent, sensitivity of 73.9 percent, and accuracy of 87.7 percent.
  • Group 2 scored a 91.6 percent specificity, sensitivity of 82.4 percent, and accuracy of 87.7 percent.
  • Group 3 achieved a specificity of 93 percent, sensitivity of 76.8 percent and accuracy of 87.8 percent.

“Prior to initiation of NAC, it is feasible to utilize CNN to predict response using a baseline MRI tumor dataset,” Ha et al. wrote.

Although the authors found their CNN to be accurate, they noted a larger dataset could improve the model and ultimately help move it toward clinical implementation.

“Our early prediction model of treatment response has the potential to impact clinical management in patients with locally advanced breast cancer, including the opportunity to direct appropriate therapy in non-responders, minimize toxicity from ineffective therapies, and facilitate the upfront use of novel targeted treatment in the neoadjuvant setting,” the authors concluded.