Investigation of the effectiveness of artificial learning methods in predicting axillary lymph node metastasis in breast cancer

Axillary lymph node involvement is one of the most significant prognostic factor of survival in women with breast cancer. Therefore, axillary area treatment options should be well determined. However, sentinel lymph node sampling to reduce the morbidity of axilla dissection may have consequences that may adversely affect quality of life. Foreseeing the presence of axillary lymph node metastases will be of great benefit in planning the treatment. In this study, the effectiveness of artificial learning methods was evaluated in the prediction of axillary lymph node metastasis. The anonymous data of 489 patient with history of full axillary dissection were collected, and the efficacy of artificial learning methods was tested. These artificial learning methods were developed with logistic regression, "k-nearest neighbour", "support vector machine", "decision tree" and "random forest" algorithms by using scikit-learn libraries. Tumor diameter, presence of lymphatic invasion and Ki-67 index showed higher correlation with axillary lymph node metastasis than other parameters in verise; instead of the categorical value of tumor grade, the total value of its components increased the correlation; and statistically significant correlation was found between preoperative tumor grade evaluations pathology reports and postoperative pathology reports. It was found that the neoadjuvant status of artificial learning methods was not sufficient in predicting the presence of axillary lymph node metastasis and that the neoadjuvant treatment status of the patient was effective on the effectiveness of the models and the addition of Ki-67 index as a variable in the dataset significantly increased the effectiveness of machine learning algorithms. Underlying reasons for the low correlation between preoperative and postoperative pathology reports must be further evaluated. Cumulative total of components of tumor grade is better correlated with axillary lymph node metastasis compared to categorical value of the tumor grade. Lymphatic invasion is the more closely corrected with axillary lymph node metastasis, and must be reported routinely and in a scalar format. Effectiveness of machine learning algorithms can be improved by including imaging and demographic data to dataset. Key words: breast cancer, axillary lymph node metastasis, artificial learning, tumor size, lymphatic invasion, Ki-67.