Texture analysis with 3.0T MRI for Association of Response to Neoadjuvant Chemotherapy in Breast Can

We investigated whether mathematical modeling with the use of texture features from MRI are associated with pathologic complete response (pCR) to neoadjuvant chemotherapy (NAC) in breast cancer. A total of 136 women who underwent NAC and subsequent surgery for breast cancer between January 2012 and August 2017. They were monitored with 3.0T MRI before (pretreatment) and after 3 or 4 cycles of NAC (midtreatment). Texture analysis was performed at pre- and midtreatment MRI with T2-weighted, contrast-enhanced T1-weighted, diffusion-weighted image (DWI) and apparent diffusion coefficient (ADC) maps by using a commercial software. A random forest method was applied to build a predictive model to classify pCR responders with texture parameters. Diagnostic performance for predicting pCR was assessed and compared with other six machine learning classifiers including adaptive boosting, decision tree, k-nearest neighbor, linear support vector machine, Naïve Bayes and linear discriminant analysis by using the Wald test and DeLong method.

Result 1. Texture features of contrast-enhanced T1-weighted MRI at midtreatment showed the highest diagnostic performance (AUC, 0.82) for predicting pCR among pre- and midtreatment MRI with T2-weighted, contrast-enhanced T1-weighted, diffusion-weighted image and apparent diffusion coefficient maps.

The Result 2. The random forest model (AUC, 0.82) has better diagnostic performance for showing association with complete pathologic response than other six machine learning classifier (AUC, Adaboost, 0.76; DT, 0.70; kNN, 0.80; LinSVM, 0.75; NB, 0.74; LDA, 0.79) on contrast-enhanced T1-weighted MRI at midtreatment.

참고 url: https://doi.org/10.1148/radiol.2019182718