Texture analysis based on contrast-enhanced MRI can predict treatment response to neoadjuvant chemotherapy of breast cancer

Abstract

Objective: To investigate whether texture analysis based on contrast-enhanced MRI can predict pathological complete response of locally advanced breast cancer undergoing neoadjuvant chemotherapy(NAC). Methods: Forty-seven patients with breast cancer undergone neoadjuvant chemotherapy from January 2015 to February 2016 were divided into pathological complete response (pCR) group or non-pathological complete response (non-pCR) group based on surgical pathology. Their parameters of texture analysis based on MRI before neoadjuvant chemotherapy and after 2 cycles of treatment were analyzed. Parameters(Energy, Entropy, Inertia, Correlation, Inverse Difference Moment)before and after 2 cycles of NAC between pCR and non-pCR groups were compared using Student t or Wilcoxon rank sum test. The diagnostic performance of different parameters was judged by the receiver-operating characteristic (ROC) curve analysis. Results: The post-NAC value was significantly different from that of pre-NAC (all P<0.05). Pre-treatment parameters (Energy, Entropy, Inertia, Correlation, Inverse Difference Moment) were 78.58×10(-5)(55.64×10(-5), 135.23×10(-5)), 10.06 ± 1.02, 7 993.91±2 428.10, (4.76±0.99) ×10(-5) and (18.10±4.13) ×10(-3) in pCR group, and 76.84×10(-5) (48.68×10(-5), 154.15×10(-5)), 10.28±1.26, 7 184.77 (4 938.03, 9 974.04), (5.21±2.01) ×10(-5) and (17.68±5.87) ×10(-3) in non-pCR group. No significant difference was found between both groups. (P>0.05 for all). At the end of the second cycle of NAC, parameters(Energy, Entropy, Inertia, Correlation, Inverse Difference Moment) were (542.11±361.04) ×10(-5,) 7.95±1.28, 16 765.08±97 06.56, (0.43±0.07) ×10(-5,) and (12.18±9.82) ×10(-3) in pCR group, and 133.00×10(-5) (79.80×10(-5,) 239.00×10(-5)), 9.29±1.46, 7 916.64(6 418.89, 10 934.40), (0.38±0.08) ×10(-5) and (14.80±5.06) ×10(-3) in non-pCR group. At the end of the second cycle of NAC, there was significant difference in the parameters (Energy, Entropy, Inertia, Correlation) and Δparameters (ΔEnergy, ΔEntropy, ΔInertia, ΔInverse Difference Moment) between both groups (P<0.05 for all). The area under curve (AUC) of post-treatment ΔEntropy was 0.81, which was the largest one among parameters. Sensitivity of ΔEntropy for predicting pCR was 75.0% and specificity was 85.7%, respectively. Conclusion: Texture analysis based on dynamic contrast-enhanced MRI can predict early treatment response in primary breast cancer.