Globally-Aware Multiple Instance Classifier for Breast Cancer Screening

Shen, Yiqiu; Wu, Nan; Phang, Jason; Park, Jungkyu; Kim, S. Gene; Moy, Linda; Cho, Kyunghyun; Geras, Krzysztof J.

doi:10.1007/978-3-030-32692-0_3

Cited by 33 publications

(56 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In [20] attention weights are learned over the input image to sample a small informative subset for the downstream per-image classification task. A related approach was introduced in [21] based on the use of saliency maps for the task of breast cancer screening with high-resolution mammography images. However, the optimisation goal of learning a spatially downsampling strategy for per-image classification is different to the per-pixel segmentation task.…”

Section: Sampling As Pre-processing For Image Classificationmentioning

confidence: 99%

Learning to Downsample for Segmentation of Ultra-High Resolution Images

Chen¹,

Tanno²,

Mertzanidou³

et al. 2021

Preprint

View full text Add to dashboard Cite

Semantic Segmentation with Deformed Downsampling. We propose a method for learning to downsample ultra high-resolution images (top-row) that reflects the importance of each location. The aim is to preserve information that guides segmentation and ignore image content that does not contribute thereby enabling low-cost semantic segmentation on low-resolution downsampled images. This strategy helps identify small regions such as the people in the figure; see masked ground truth and predictions in middle-row. We introduce the deformation module, a learnable "downsampler", that can be jointly trained with the segmentation module, to adapt sampling density according to importance at each location (white dots on the images). The deformation module is lightweight and can be flexibly integrated into existing semantic segmentation architectures. Compared to baseline standard uniform downsampling (referred to as "Uniform") our method (referred to as "Ours") samples more densely at important semantic regions and leads to more accurate segmentation (see bottom-row).

show abstract

Section: Sampling As Pre-processing For Image Classificationmentioning

confidence: 99%

Learning to Downsample for Segmentation of Ultra-High Resolution Images

Chen¹,

Tanno²,

Mertzanidou³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…For example, mammography images have a much higher resolution (∼ 10 7 pixels) than natural images (∼ 10 5 pixels) in most benchmark datasets, such as ImageNet (Deng et al, 2009). Because of this, when applied to medical images, CNNs often aggressively downsample the input image (Shen et al, 2019(Shen et al, , 2021 to accommodate GPU memory constraints, making the resulting localization too coarse. This is a crucial limitation for many medical diagnosis tasks, where regions of interest (ROIs) are often small (e.g.…”

Section: Introductionmentioning

confidence: 99%

“…The majority of these works focus on images that have relatively low resolution, that is, 512 × 512 pixels or less. Only a few works have considered higher resolution images, which are standard in some imaging procedures such as screening mammography (Shen et al, 2019(Shen et al, , 2021.…”

Section: Introductionmentioning

confidence: 99%

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

Liu,

Shen,

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

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.

show abstract

“…Breast Cancer has become the main reason behind the death of a lot of women all around the world.The main reason for the death of women by this disease is the process by which it is diagnosed 1 .…”

Section: Introductionmentioning

confidence: 99%

“…The technology has become a major part of our lifestyles still we are lacking behind diagnosing this critical disease in early stages [1]. As the disease is not diagnosed in early stages, therefore, the mammography rate has been increased for a particular age group of concerned women [2].Breast Cancer is curable and life can be saved if it is diagnosed in early stages.…”

Section: Introductionmentioning

confidence: 99%

Machine Learning Techniques and Extreme Learning Machine for Early Breast Cancer Prediction

Gupta*¹,

Gill²

2020

IJITEE

View full text Add to dashboard Cite

Breast Cancer is one of the most deadly disease and most of the women are infected by this vital disease in many parts of the world. Medical tests conducted in hospitals for determining the disease are very much expensive as well as timeconsuming. The problem can be resolved by diagnosing the problem in early spam of time and by providing results with more accuracy.In this paper, different machine learning and neural network algorithmhave been studied and compared to predict cancer in early stages so that life can be saved. The dataset available publically for Breast Cancer has been used. Different algorithms compared includeSupport Vector Machine Classification (SVM), K-Nearest NeighborClassification (KNN), Decision tree Classification (DT), Random Forest Classification (RF) and Extreme Learning Machine (ELM).All are compared on the basis of Accuracy and processing time are considered as the parameters for comparing analysis. The results reveal that extreme learning machine comes to be the better algorithm.

show abstract

Globally-Aware Multiple Instance Classifier for Breast Cancer Screening

Cited by 33 publications

References 15 publications

Learning to Downsample for Segmentation of Ultra-High Resolution Images

Learning to Downsample for Segmentation of Ultra-High Resolution Images

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

Machine Learning Techniques and Extreme Learning Machine for Early Breast Cancer Prediction

Contact Info

Product

Resources

About