Diagnosis of breast cancer lesion using ultrasound images, elastography, and Ki-67 protein cell proliferation index

Today, using elastography and ultrasound images is the best method for diagnosing breast cancer for dense tissues, especially for women under 30 years old, which is used to detect the exact border of masses. Besides, using quantitative microscopic criteria that are less tasteful seems to be useful in predicting the behavior of the tumor and its prognosis. Ki-67 is an antigen corresponding to a nuclear non-histone protein produced by cells in proliferative phases. In this article, ultrasound and elastography images of patients were collected, and breast masses were identified. The proposed algorithm includes pre-processing, feature extraction, and classification. To remove the speckle noise, two pre-processing steps are used, and after segmenting each data with its appropriate color channel, statistical features and features based on the morphology of suspicious areas are extracted. Also, sections of paraffin blocks of samples fixed in formalin were prepared and stained by immunohistochemical staining with Ki-67 monoclonal antibody, and the cell proliferation index was determined in the prepared slides. The relationship between Ki-67 positivity and microscopic grade was studied. The feature extraction results show that elastography is chosen as a more appropriate method than ultrasound due to the separation in terms of color channels. The most appropriate proposed combined methods, namely RBF-Kmeans, MLP-SCG, and RBF-SOM, have been used to classify features. The combined MLP-SCG classifier with an average accuracy of 96% and an average of 98% has improved significantly compared to other methods.

Automated summary

Today, using elastography and ultrasound images is the best method for diagnosing breast cancer for dense tissues, especially for women under 30 years old, which is used to detect the exact border of masses. Besides, using quantitative microscopic criteria that are less tasteful seems to be useful in predicting the behavior of the tumor and its prognosis. The proposed algorithm includes pre-processing, feature extraction, and classification, aiming to improve the accuracy of breast cancer diagnosis.