Image processing based lesion detection in breast cancer

The aim of this study is to determine the breast tumor site with high accuracy and low error rate. In this study, K-mean clustering and herbaceous thresholding methods were used in 8 cancer images taken from the TCIA (The Cancer Imaging Archive) data bank. For the clustering process, TPR (True Positive Rate) 0.89, FPR (False Positive Rate) 0.14, TNR (True Negative Rate) 0.86, FNR (False Negative Rate) 0.10, similarity 0.67, accuracy 0.87, sensitivity 0.89, sensitivity 0.86, specificity 0.87, F score 0.87 were found, respectively. TPR (True Positive Rate) 0.84, FPR (False Positive Rate) 0.12, TNR (True Negative Rate) 0.89, FNR (False Negative Rate) 0.14, similarity 0.73, accuracy 0.84, sensitivity 0.84, sensitivity 0.86, specificity 0.87, F score 0.84 were calculated for Otsu. After that, in addition to 8 images, 646 images were taken from the Kaggle data bank, 504 were October for training and 142 were reserved for testing. Region detection was performed with U-Net artificial neural network architecture. The tumor region obtained as a result of network training was compared with the tumor region in the marked reference image.With an Epoch value of 100, the test result was achieved with an accuracy of 96% with a loss rate of 4%. As a result, it has been determined that these methods can be used for tumor detection, and the artificial neural network model has found the highest accuracy.