2020
DOI: 10.48550/arxiv.2002.07613
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

An interpretable classifier for high-resolution breast cancer screening images utilizing weakly supervised localization

Abstract: Medical images differ from natural images in significantly higher resolutions and smaller regions of interest. Because of these differences, neural network architectures that work well for natural images might not be applicable to medical image analysis. In this work, we extend the globally-aware multiple instance classifier, a framework we proposed to address these unique properties of medical images. This model first uses a low-capacity, yet memory-efficient, network on the whole image to identify the most i… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
4
1

Citation Types

0
6
1

Year Published

2020
2020
2021
2021

Publication Types

Select...
1
1

Relationship

1
1

Authors

Journals

citations
Cited by 2 publications
(7 citation statements)
references
References 90 publications
0
6
1
Order By: Relevance
“…In addition, we evaluate the models' performance on a challenging population which consists only of cases that are difficult to diagnose and the radiologist requested a biopsy for. This further differentiates our work from previous works [16,30,17,13,18] and makes our results not directly comparable to theirs. This is because these methods were developed and evaluated for the screening population, which contains a lot of negative cases not requiring biopsy, which can inflate their evaluation metrics [16].…”
Section: Introductioncontrasting
confidence: 82%
See 4 more Smart Citations
“…In addition, we evaluate the models' performance on a challenging population which consists only of cases that are difficult to diagnose and the radiologist requested a biopsy for. This further differentiates our work from previous works [16,30,17,13,18] and makes our results not directly comparable to theirs. This is because these methods were developed and evaluated for the screening population, which contains a lot of negative cases not requiring biopsy, which can inflate their evaluation metrics [16].…”
Section: Introductioncontrasting
confidence: 82%
“…We refer to this network as the "context network." We use Globally-Aware Multiple Instance Classifier [30,17] as the context network, which is explicitly designed to provide interpretability by highlighting the most informative regions of the input images. To be more precise, the feature maps obtained after the last residual block of the context network are transformed by a 1×1 convolutional layer with sigmoid activation into two saliency maps, denoted as S m ∈ [0, 1] 46,30 and S b ∈ [0, 1] 46, 30 .…”
Section: The Proposed Methodsmentioning
confidence: 99%
See 3 more Smart Citations