| Summary: | The breast cancer screening process using mammograms with the aid of a deep learning-based computer-aided (CAD) system can decrease the cause of the human error, improve patient monitoring and diagnosis, reduce false positive rates, and improve patient treatment options and care. However, system development based on only automated feature maps without the input knowledge from the radiologists causes concern of creating a system that is not easily tuned to be adjusted with more detailed cancer features according to the latest expert radiologist's views in the future. By incorporating the traditional machine learning (ML) method in feature mapping and classification, dual-feature training based on both mammogram images and handcrafted features is used as the input for a hybridized deep learning Convolutional Neural Network (CNN) with the Support Vector Machine (SVM) method for classification of mass benign, malignant, and normal (fatty and fibro-glandular) tissue. Cropped input images are utilized to overcome the limitations of small input training images. The result shows an increase in performance for an overall four classes with an accuracy of 93.01%, as well as benign vs. malignant of 98.51% and fatty vs. fibroglandular of 91.33% in the system developed based on the dual-feature on the CNN and SVM-based frameworks. Including radiologist-based radiomics handcrafted features with automated mammogram image features in determining cancer mass images help create a promising CAD diagnostics performance. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
|
|---|