Biomedical mammography images classification by patches based feature engineering using deep learning with ensemble classifier
Saruchi Kukkar, Jaspreet Singh · AIP conference proceedings · 2024
Deep learning algorithms have been in use in the mammography processing business recently to help cut radiologists' cost.Breast masses are currently classified using deep learning-based techniques, including a Convolutional Neural Network (CNN).CNN-based systems have a clear advantage over machine learning-based systems when it comes to categorizing mammography images, but it does have its drawbacks.Additional difficulties are a lack of information about feature engineering, and feature analysis is not feasible for current patches of photos, which are not very distinct in low contrast mammograms.Mammography image patches have contributed to an increase in misleading information, greater computation costs, faulty patch assessments, and non-recovered patch intensity variance.Because of this, it was shown that with the help of a convolutional neural network (CNN), a CNN-based method for classifying breast masses obtained poor classification accuracy.This innovative breast mass categorization methodology, dubbed Deep Learning Feature Engineering, is used to improve accuracy on low contrast images (DFN).To characterize breast masses, this system incorporates both CNN architectures, such as VGG 16 and Resnet 50, as well as random forest boosting techniques.The proposed DFN method is also compared to current classification algorithms that utilize two publicly available datasets of mammographic photos.