Weakly-supervised High-resolution Segmentation of Mammography Images for Breast Cancer Diagnosis

Abstract

In the last few years, deep learning classifiers have shown promising results in image-based medical diagnosis. However, interpreting the outputs of these models remains a challenge. In cancer diagnosis, interpretability can be achieved by localizing the region of the input image responsible for the output, i.e. the location of a lesion. Alternatively, segmentation or detection models can be trained with pixel-wise annotations indicating the locations of malignant lesions. Unfortunately, acquiring such labels is labor-intensive and requires medical expertise. To overcome this difficulty, weaklysupervised localization can be utilized. These methods allow neural network classifiers to output saliency maps highlighting the regions of the input most relevant to the classification task (e.g. malignant lesions in mammograms) using only image-level labels (e.g. whether the patient has cancer or not) during training. When applied to high-resolution images, existing methods produce lowresolution saliency maps. This is problematic in applications in which suspicious lesions are small in relation to the image size. In this work, we introduce a novel neural network architecture to perform weakly-supervised segmentation of high-resolution images. The proposed model selects regions of interest via coarse-level localization, and then performs fine-grained segmentation of those regions. We apply this model to breast cancer diagnosis with screening mammography, and validate it on a large clinically-realistic dataset. Measured by Dice similarity score, our approach outperforms existing methods by a large margin in terms of localization performance of benign and malignant lesions, relatively improving the performance by 39.6% and 20.0%, respectively. Code and the weights of some of the models are available at https://github.com/nyukat/GLAM.

Multi-view Local Co-occurrence and Global Consistency Learning Improve Mammogram Classification Generalisation

“…Importantly, these models are not fine-tuned to any of the datasets. Testing results based on area under the receiver operating characteristic curve (AUC) show that the GMIC model [24] is about 20% better than Faster R-CNN [18] on the NYU reader study test set [31], but Faster R-CNN outperforms GMIC on the INBreast dataset [14] by 4%, and GLAM [12] reaches 85.3% on the NYU test set but only gets 61.2% on INBreast and 78.5% on CMMD [15]. Such experiments show the importance of assessing the ability of models to generalise to testing sets from different populations and with images produced by different machines, compared with the training set.…”

Métodos fracamente supervisionados para classificação de ácaros na citricultura

“…A ideia é similar aos métodos propostos nesta tese, porém, por ser treinada como um classificador ponta a ponta, a GMIC pode não ser adequada para a classificação na CPB (Subseção 6.2.7). Liu et al [112] propuseram uma arquitetura fracamente supervisionada que segue a mesma linha da GMIC, cujo fluxo é realizado em etapas, entretanto, ainda une as duas arquiteturas na etapa final, chamada mapas de ativações locaisglobais (global-local activation maps, GLAM), também no contexto de detecção de câncer de mama. Primeiramente, eles treinaram a arquitetura global, depois a congelaram e treinaram a arquitetura local com os cortes produzidos pela arquitetura global e, na última etapa, refinaram o treinamento das duas arquiteturas ao mesmo tempo.…”

“…2020 Shen et al [166] Evolução do GMIC para melhorar os resultados e melhorar os tempos de inferência. 2021 Liu et al [112] Evolução do GMIC chamada GLAM. 2021 Luo et al [122] Mapas de ativação para diminuir a influência do plano de fundo no treinamento de imagens com pragas.…”