Using a Wavelet-Based and Fine-Tuned Convolutional Neural Network for Classification of Breast Density in Mammographic Images
Eri Matsuyama, Megumi Takehara, Du‐Yih Tsai · Open Journal of Medical Imaging · 2020
Classification of breast density is significantly important during the process of breast diagnosis. The purpose of this study was to develop a useful computer-ized tool to help radiologists determine the patient’s breast density category on the mammogram. In this article, we presented a model for automatically classi-fying breast densities by employing a wavelet transform-based and fine-tuned convolutional neural network (CNN). We modified a pre-trained AlexNet model by removing the last two fully connected (FC) layers and appending two newly created layers to the remaining structure. Unlike the common CNN-based methods that use original or pre-processed images as inputs, we adopted the use of redundant wavelet coefficients at level 1 as inputs to the CNN model. Our study mainly focused on discriminating between scattered density and heterogeneously dense which are the two most difficult density cat-egories to differentiate for radiologists. The proposed system achieved 88.3% overall accuracy. In order to demonstrate the effectiveness and usefulness of the proposed method, the results obtained from a conventional fine-tuning CNN model was compared with that from the proposed method. The results demon-strate that the proposed technique is very promising to help radiologists and serve as a second eye for them to classify breast density categories in breast cancer screening.