Feasibility of predicting breast cancer Ki-67 expression using histogram features derived from time-dependent...

Through the quantification of multiple microstructural parameters, time-dependent diffusion magnetic resonance imaging (Td-dMRI) offers a novel approach for establishing urgently needed imaging biomarkers of tumor heterogeneity. This study employed Td-dMRI-derived histogram parameters to predict Ki-67 expression in breast cancer.

A total of 86 female patients with breast cancer were enrolled in this prospective study. Microstructural parameters, including mean cell diameter (d), intracellular volume fraction (f_in), intracellular diffusion coefficient (D_in), extracellular apparent diffusion coefficient (D_ex), and intracellular water pre-exchange lifetime (Tau_in), were estimated using the JOINT model based on Td-dMRI data acquired with pulsed (PGSE) and oscillating (OGSE) gradient spin-echo sequences. Two additional parameters, intracellular water exchange rate (K_in) and cellularity, were subsequently derived through calculation. The apparent diffusion coefficient (ADC) was calculated from the conventional diffusion-weighted imaging (DWI), PGSE (32 and 52 ms), and OGSE (17 and 33 Hz) sequences. Histogram features were extracted based on parameters derived from Td-dMRI. The independent t-test or Mann-Whitney U test was used to compare differences between Ki-67 groups. The dataset was divided into a training set and test set at a ratio of 7:3 for internal validation. Clinical and conventional imaging characteristics identified by univariate regression analysis (P<0.1), together with histogram features selected by least absolute shrinkage and selection operator (LASSO) regression, were used to further establish a logistic regression model in the training set. Model performance was assessed by receiver operating characteristic (ROC) curves in the training and test sets. The area under the curve (AUC), sensitivity, specificity, and accuracy were calculated.

Nine histogram parameters differed significantly between the Ki-67 expression groups; however, no significant difference in ADC_DWI was observed (P=0.859). Axillary lymph node metastasis (ALNM) was identified as an independent predictor of high Ki-67 expression (odds ratio (OR) =2.991; 95% confidence interval (CI): 0.964-9.278, P=0.058). Among the eight histogram features ultimately selected by LASSO regression, Cellularity_P10 and ADC_33Hz_Skewness contributed the most to the model. The combined model achieved an AUC of 0.890 (95% CI: 0.809-0.971), with a sensitivity, specificity, and accuracy of 89%, 65%, and 80%, respectively, in the training set, and an AUC of 0.870 (95% CI: 0.712-1), with a sensitivity, specificity, and accuracy of 80%, 50%, and 73%, respectively in the test set.

Td-dMRI-derived histogram features show promise for predicting Ki-67 expression in breast cancer and may complement conventional imaging.