A Breast Density Classification System for Mammography Considering Reliability Issues in Deep Learning

In a convolutional neural network (CNN) classification model for diagnosing medical images, transparency and interpretability of the model’s behavior are crucial in addition to high classification accuracy, and it is highly important to explicitly demonstrate them. In this study, we constructed an interpretable CNN-based model for breast density classification using spectral information from mammograms. We evaluated whether the model’s prediction scores provided reliable probability values using a reliability diagram and visualized the basis for the final prediction. In constructing the classification model, we modified ResNet50 and introduced algorithms for extracting and inputting image spectra, visualizing network behavior, and quantifying prediction ambiguity. From the experimental results, our proposed model demonstrated not only high classification accuracy but also higher reliability and interpretability compared to the conventional CNN models that use pixel information from images. Furthermore, our proposed model can detect misclassified data and indicate explicit basis for prediction. The results demonstrated the effectiveness and usefulness of our proposed model from the perspective of credibility and transparency.

Using a Wavelet-Based and Fine-Tuned Convolutional Neural Network for Classification of Breast Density in Mammographic Images

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.