Application of machine learning methods for prediction of breast cancer recurrence: A study on pathology image features

Introduction & Aim: Breast cancer recurrence is dependent to many biological and clinical features, but there are limited studies based on pathology images for prediction of recurrence. The main purpose of present study was to predict breast cancer recurrence using machine learning approaches. Methods: This retrospective study carried out on pathology image dataset including 138 non-recurrences and 46 recurrences breast cancer patients. The recurrence was assessed beyond 24 months based on clinicalendpoints. 33 different image features including intensity, texture and shape extracted from fine needle aspiration (FNA) pathology images and were used as input for machine learning methods. Logistic regression model was used as machine learning approach. Principal component analysis (PCA) and 10-fold cross validation were used for dimension reduction and model validation respectively. Results: PCA technique reduced our features from 33 to 13. The tested regression model showed breast cancer recurrence prediction with AUC = 0.75, accuracy = 0.80, sensitivity = 0.80 and precision = 0.78. Conclusion: Machine learning methods can be used as feasible and easy to use approaches for breast cancer prognosis. Using these methods, we predict breast cancer recurrence with highest accuracy and sensitivity. In the other hand these methods, clinicians can save their times and costs and better prediction power.