Detecting And Classifying Micro-calcifications In Mammography Images Through Deep Learning

Mammography is an important technique for early screening and diagnosis of breast cancer.Both malignant and benign breast lesions are accompanied by micro-calcifications,and different calcification morphological characteristics are formed.Many scholars have proposed the diagnosis of breast calcification based on artificial extraction features and traditional machine learning algorithm.However,these methods generate high false positive rate and the processing is in the mode of semi-automatic diagnosis,with low detection efficiency.Therefore,this paper proposes a full-automatic aiding diagnosis method,for the detection and classification of micro-calcifications images based on deep learning.The main work is as follows:This method is divided into two important parts,breast micro-calcifications detection and classification.In the detection of breast calcification,the mammography was uniformly cropped to 30(5 × 6)sub-images because of its high resolution.The calcification detection model is used to detect the sub-image.Six calcification categories and their corresponding positions can be detected after calculation.After sub-images are merged,several categories of calcification can be selected and displayed.Depending on the calcification detection module,calcification clustering is carried out according to the calcification priority to synthesize the calcification focused regional images,and then the regional images are tested in the classification model.After calculation,the benign and malignant categories of mammography can be classified.In this study,448 benign micro-calcifications images and 544 malignant micro-calcifications images were used,and each image had corresponding pathological images as the gold standard.After experimental statistics,the average accuracy rate of the six micro-calcifications types was 89.70%,and the accuracy rate of benign and malignant was 89.52% for the M-DenseNet model.Obviously,detection and classification model based on deep learning have achieved a high micro-calcifications recognition accuracy.In summary,the combination of detection model and classification model can realize the automatic breast micro-calcifications,thus aiding diagnosis by showing a high accuracy in recognition.Therefore,the method of detection and classification of micro-calcifications image based on deep learning provides an effective aided function for early screening and diagnosis,reducing risks and improving efficiency...