| Summary: | Identifying breast cancer at an early stage is an important part of determining a suitable treatment plan but is often challenging due to the image quality. Digital Breast Tomosynthesis (DBT) is a method that extends digital mammography to detect breast cancer beyond areas of density. However, the series of projection images of the breast from various angles of the DBT system X-ray source results in blurry and low-contrast effect images. Due to this issue, the identification of abnormalities among DBT series is tedious and time-consuming. Therefore, there is an urgent need for automatic image quality analysis techniques on DBT images. This study proposed a Deep Learning (DL) based approach for blur image detection in DBT series of images. A Convolutional Neural Network (CNN) was constructed from scratch and then used in hybrid with a support vector machine (SVM) classifier to detect blur in DBT images. The performance of the obtained model was attempted to be improved by training with various deep-learning optimizers including Sgdm, Adam, and RMSProp. For a comprehensive evaluation, training accuracy, validation accuracy, F1-score, and the number of epochs (required to converge training and validation plots) were compared. From the experiments, the proposed constructed CNN architecture trained with the Adam optimizer attained the highest validation accuracy and F1-score of 98.04% and 0.9796 respectively at 50 epochs which is comparatively better than other optimizers. The finding of this study shows that DL optimizer evaluation is necessary to get the optimal performance of the CNN model. © 2023 IEEE.
|
|---|