Some recent studies have described deep convolutional neural networks to diagnose breast cancer in mammograms with similar or even superior performance to that of human experts. One of the best techniques does two transfer learnings: the first uses a model trained on natural images to create a "patch classifier" that categorizes small subimages; the second uses the patch classifier to scan the whole mammogram and create the "single-view whole-image classifier". We propose to make a third transfer learning to obtain a "two-view classifier" to use the two mammographic views: bilateral craniocaudal and mediolateral oblique. We use EfficientNet as the basis of our model. We "end-to-end" train the entire system using CBIS-DDSM dataset. To ensure statistical robustness, we test our system twice using: (a) 5-fold cross validation; and (b) the original training/test division of the dataset. Our technique reached an AUC of 0.934 using 5-fold cross validation (accuracy, sensitivity and specificity are 85.13% at the equal error rate point of ROC). Using the original dataset division, our technique achieved an AUC of 0.8483, as far as we know the highest reported AUC for this problem, although the subtle differences in the testing conditions of each work do not allow for an accurate comparison. The inference code and model are available at https://github.com/dpetrini/two-views-classifier.INDEX TERMS Breast cancer diagnosis, deep learning, convolutional neural network, mammogram, transfer learning.

e13553 Background: Mammography interpretation presents some challenges however, better technological approaches have allowed increased accuracy in cancer diagnosis and nowadays, radiologists sensitivity and specificity for mammography screening vary from 84.5 to 90.6 and 89.7 to 92.0%, respectively. Since its introduction in breast image analysis, artificial intelligence (AI) has rapidly improved and deep learning methods are gaining relevance as a companion tool to radiologists. Thus, the aim of this systematic review and meta analysis was to evaluate the sensitivity and specificity of AI deep learning algorithms and radiologists for breast cancer detection through mammography. Methods: A systematic review was performed using PubMed and the words: deep learning or convolutional neural network and mammography or mammogram, from January 2015 to October 2020. All titles and abstracts were doubly checked; duplicate studies and studies in languages other than English were excluded. The remaining complete studies were doubly assessed and those with specificity and sensibility information had data collected. For the meta analysis, studies reporting specificity, sensitivity and confidence intervals were selected. Heterogeneity measures were calculated using Cochran Q test (chi-square test) and the I2 (percentage of variation). Sensitivity and specificity and 95% confidence intervals (CI) values were calculated, using Stata/MP 14.0 for Windows. Results: Among 223 studies, 66 were selected for full paper analysis and 24 were selected for data extraction. Subsequently, only papers evaluating sensitivity, especificity, CI and/or AUC were analyzed. Eleven studies compared AUC using AI with another method and for these studies, a differential AUC was calculated, however no differences were observed: AI vs Reader (n = 3; p = 0.109); AI vs AI (n = 5; p = 0.225); AI vs AI + reader (n = 2; p = 0.180); AI + Reader vs reader (n = 2; p = 0.655); AI vs reader (n > 1) (n = 3; p = 0.102). Some studies had more than one comparison. A meta analysis was performed to evaluate sensitivity and specificity of the methods. Five studies were included in this analysis and a great heterogeneity among them was observed. There were studies evaluating more than one AI algorithm and studies comparing AI with readers alone or in combination with AI. Sensitivity for AI; AI + reader; reader alone, were 76.08; 84.02; 80.91, respectively. Specificity for AI; AI + reader; reader alone, were 96.62; 85.67; 84.89, respectively. Results are shown in the table. Conclusions: Although recent improvements in AI algorithms for breast cancer screening, a delta AUC between comparisons of AI algorithms and readers was not observed.[Table: see text]

Background:Early computer-aided detection systems for mammography have failed to improve the performance of radiologists. With the remarkable success of deep learning, some recent studies have described computer systems with similar or even superior performance to that of human experts. Among them, Shen et al. (Nature Sci. Rep., 2019) present a promising “end-to-end” training approach. Instead of training a convolutional net with whole mammograms, they first train a “patch classifier” that recognizes lesions in small subimages. Then, they generalize the patch classifier to “whole image classifier” using the property of fully convolutional networks and the end-to-end approach. Using this strategy, the authors have obtained a per-image AUC of 0.87 [0.84, 0.90] in the CBIS-DDSM dataset. Standard mammography consists of two views for each breast: bilateral craniocaudal (CC) and mediolateral oblique (MLO). The algorithm proposed by Shen et al. processes only single-view mammography. We extend their work, presenting the end-to-end training of convolutional net for two-view mammography. Methods:First, we reproduced Shen et al.'s work, using the CBIS-DDSM dataset. We trained a ResNet50-based net for classifying patches with 224x224 pixels using segmented lesions. Then, the weights of the patch classifier were transferred to the whole image single-view classifier, obtained by removing the dense layers from the patch classifier and stacking one ResNet block at the top. This single-view classifier was trained using full images from the same dataset. Trying to replicate Shen et al.'s work, we obtained an AUC of 0.8524±0.0560, less than 0.87 reported in the original paper. We attribute this worsening to the fact that we are using only 2260 images with two views, instead of 2478 images from the original work. Finally, we built the two-view classifier that receives CC and MLO views as input. This classifier has inside two copies of the patch classifier, loaded with the weights from the single-view classifier. The features extracted by the two patch classifiers are concatenated and submitted to the ResNet block. The two-view classifier is end-to-end trained using full images, refining all its weights, including those inside the two patch classifiers. Results:The two-view classifier yielded an AUC of 0.9199±0.0623 in 5-fold cross-validation to classify mammographies into malignant/non-malignant, using single-model and without test-time data augmentation. This is better than the Shen et al.'s AUC (0.87), our single-view AUC (0.85). Zhang et al. (Plos One, 2020) present another two-view algorithm (without end-to-end training) with AUC of 0.95. However, this work cannot directly be compared with ours, as it was tested on a different set of images. Conclusions:We presented end-to-end training of convolutional net for two-view mammography. Our system's AUC was 0.92, better than the 0.87 obtained by the previous single-view system. Citation Format: Daniel G. Petrini, Carlos Shimizu, Gabriel V. Valente, Guilherme Folgueira, Guilherme A. Novaes, Maria L. Katayama, Pedro Serio, Rosimeire A. Roela, Tatiana C. Tucunduva, Maria Aparecida A. Folgueira, Hae Y. Kim. End-to-end training of convolutional network for breast cancer detection in two-view mammography [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 183.

e14068 Background: The interpretation of the mammography is challenging, especially in young women, who have dense breasts. Artificial intelligence (AI) promises to improve breast cancer detection; however these systems should be tested on different datasets. Our aim is to evaluate the performance of a publicly available deep convolutional neural network, developed by Wu et al. (IEEE Trans. Med. Imaging, 2019), applied to mammograms of young women. Methods: The test dataset consisted of mammograms obtained on a single occasion from 135 young women (up to 40 years old) on a Siemens mammography system. Each exam consisted of 4 full-field digital mammography images and had two labels (left malignant and right malignant). Mammograms were analyzed by a single mammography trained radiologist, using BI-RADS reporting tool. Among 270 labels, 170 were malignant and 100 were non-malignant. We used the program developed by Wu et al. that, according to the authors, presents AUC of 0.895 for the general population. As a preliminary test, we ran this program in a publicly available dataset named INbreast and obtained AUC of 0.8708, very close to the result reported by the authors. Results: We applied the program to our dataset of young women and obtained AUC of 0.876. We computed its standard error, obtaining 0.0290. At equal error rate point of the ROC curve, specificity and sensitivity are both 0.774. With this result we conclude that, at least for our dataset, cancer detection in young women is not substantially more difficult than in general population for an AI system. We fine-tuned the weights of the original network to the population of young women using transfer learning and obtained a slight improvement in AUC: 0.9018±0.0528, where the mean and the standard error were obtained using 5-fold cross validation. As the improvement was small and the standard errors are large, we would have to test on a larger test set to ensure that the observed improvement is real. Conclusions: We conclude, based on the experimental data, that there is no substantial degradation in accuracy when a mammogram screening program for general population is used for young women. We also conclude that it seems to be possible to obtain a slight improvement in accuracy by fine-tuning the network for the population of young women.