PURPOSE: The aim of this study was to develop a combined statistical model using both clinicopathological factors and texture parameters from F-FDG PET/CT to predict responses to neoadjuvant chemotherapy in patients with breast cancer. MATERIALS AND METHODS: A total of 435 patients with breast cancer were retrospectively enrolled. Clinical and pathological data were obtained from electronic medical records. Texture parameters were extracted from pretreatment FDG PET/CT images. The end point was pathological complete response, defined as the absence of residual disease or the presence of residual ductal carcinoma in situ without residual lymph node metastasis. Multivariable logistic regression modeling was performed using clinicopathological factors and texture parameters as covariates. RESULTS: In the multivariable logistic regression model, various factors and parameters, including HER2, histological grade or Ki-67, gradient skewness, gradient kurtosis, contrast, difference variance, angular second moment, and inverse difference moment, were selected as significant prognostic variables. The predictive power of the multivariable logistic regression model incorporating both clinicopathological factors and texture parameters was significantly higher than that of a model with only clinicopathological factors (P = 0.0067). In subgroup analysis, texture parameters, including gradient skewness and gradient kurtosis, were selected as independent prognostic factors in the HER2-negative group. CONCLUSIONS: A combined statistical model was successfully generated using both clinicopathological factors and texture parameters to predict the response to neoadjuvant chemotherapy. Results suggest that addition of texture parameters from FDG PET/CT can provide more information regarding treatment response prediction compared with clinicopathological factors alone.
PURPOSE: The aim of this study was to develop a combined statistical model using both clinicopathological factors and texture parameters from F-FDG PET/CT to predict responses to neoadjuvant chemotherapy in patients with breast cancer. MATERIALS AND METHODS: A total of 435 patients with breast cancer were retrospectively enrolled. Clinical and pathological data were obtained from electronic medical records. Texture parameters were extracted from pretreatment FDG PET/CT images. The end point was pathological complete response, defined as the absence of residual disease or the presence of residual ductal carcinoma in situ without residual lymph node metastasis. Multivariable logistic regression modeling was performed using clinicopathological factors and texture parameters as covariates. RESULTS: In the multivariable logistic regression model, various factors and parameters, including HER2, histological grade or Ki-67, gradient skewness, gradient kurtosis, contrast, difference variance, angular second moment, and inverse difference moment, were selected as significant prognostic variables. The predictive power of the multivariable logistic regression model incorporating both clinicopathological factors and texture parameters was significantly higher than that of a model with only clinicopathological factors (P = 0.0067). In subgroup analysis, texture parameters, including gradient skewness and gradient kurtosis, were selected as independent prognostic factors in the HER2-negative group. CONCLUSIONS: A combined statistical model was successfully generated using both clinicopathological factors and texture parameters to predict the response to neoadjuvant chemotherapy. Results suggest that addition of texture parameters from FDG PET/CT can provide more information regarding treatment response prediction compared with clinicopathological factors alone.
Authors: David K Woolf; Sonia P Li; Simone Detre; Alison Liu; Andrew Gogbashian; Ian C Simcock; James Stirling; Michael Kosmin; Gary J Cook; Muhammad Siddique; Mitch Dowsett; Andreas Makris; Vicky Goh Journal: Biomark Cancer Date: 2019-06-04