The contribution of radiological texture analysis to preoperative prediction of axillary lymph node metastasis in breast malign tumors

Purpose In this study, where the images of breast masses in patients who have undergone breast MRI with the diagnosis of breast cancer were examined, it was aimed to investigate the contribution of the data obtained to the preoperative prediction of axillary lymph node metastasis of breast cancer. In addition, the study aimed to create a model through machine learning. Material and Method In the study, the records and images of 115 patients who underwent MRI examinations with a preliminary diagnosis of breast tumor between January 2015 and December 2020 were evaluated retrospectively. Patients with missing file records or pathology results, diagnosed with non-breast cancer, and whose imaging features were not suitable for examination were excluded from the study. Images of 86 patients included in the study were transferred to open-source software LIFEx, where pixel size adjustment, grey level discretization, and grey level normalization processes were applied. Then, the relevant region (ROI) was determined from the axial section where the lesion was the widest, and 38 texture features were calculated for each patient over this ROI. The diagnostic decision-making properties of texture analysis parameters in predicting lymph node metastasis were examined by receiver operating characteristic (ROC) analysis. In addition, these data were transferred to the MATLAB R2020a (Math-Works, Natick, Massachusetts) program and the probability of accurate prediction of the model was evaluated through machine learning. Kolmogorov-Smirnov tests, Mann-Whitney U test, Kruskal Wallis test, Chi-Square test Spearman correlation, ROC analysis, and logistic regression statistical tests were used in the study. Results In the study, no significant correlation was found between the presence of lymph node metastases and the histogram features of primary breast tumor tissue. However, when the second-order statistical features and volume features were evaluated, a significant relationship was found between the 4 features (GRLLM_SRLGE, GLRLM_LRLGE, GLCM_Entropy_log2 and Shape_Volume_vx) and the presence of ALN metastasis. Although SHAPE_Volume_vx remained significant after multiple logistic regression analysis, other parameters were not significant. By transferring these 4 texture features to the MATLAB R2020a program, the probability of accurate prediction in the Fine Gaussian SVM machine learning model is 72.1%. (AUC:0.63). Conclusion In the study, it was found out that the texture analysis features obtained from preoperative breast MRI were statistically significant with the presence of ALN metastases. In order to verify the validity of the data obtained by texture analysis methods in breast malignant tumors, to be the gold standard method and to obtain reference values, multicenter research studies planned with standardized methods in large patient groups are needed. Key Words: Breast cancer, axillary lymph node metastasis, texture analysis, magnetic resonance imaging, preoperative prediction