| Summary: | Digital breast tomosynthesis (DBT) is a method that extends digital mammography by capturing pictures of the breast from various angles of the x-ray source. DBT's angular sampling range is severely limited due to hardware constraints, resulting in severely limited angular artefacts such as blurring and low contrast effects in the reconstructed images. Unwanted artefacts like blurry artefacts can substantially obscure the cancer site, particularly in exceptionally thick fibro glandular breast tissue, and reduce diagnostic accuracy. Due to the blurry artefact problem, it is essential to develop methods for analyzing the blur distortion of DBT-obtained images for diagnostic reasons. This chapter describes a hybrid convolutional neural network-support vector machine (CNN-SVM) strategy extracting robust hierarchical features from images using CNN before passing the images to an SVM classifier for classifier boosting to categorize DBT images into two classes: blur or non-blur images. To make the prediction invariance of image scaling and rotation more robust, a variety of data augmentation strategies are examined. The suggested tool was evaluated using the metrics of overall accuracy, recall, precision and processing time. The findings demonstrate that the combined CNN and SVM model outperforms standard feature models with an accuracy of 0.97 and an area under the curve (AUC) of the receiver operating characteristic (ROC) of 0.9998, as well as numerous classical deep CNN models. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
|
|---|