Does Medical Imaging learn different Convolution Filters Paul Gavrikov1Janis Keuper12

2025-05-03 0 0 2.27MB 5 页 10玖币

侵权投诉

Does Medical Imaging learn different Convolution

Filters?

Paul Gavrikov1Janis Keuper1,2

1IMLA, Offenburg University, Germany

2Fraunhofer ITWM, Kaiserslautern, Germany

{paul.gavrikov, janis.keuper}@hs-offenburg.de

Abstract

Recent work has investigated the distributions of learned convolution ﬁlters through

a large-scale study containing hundreds of heterogeneous image models. Surpris-

ingly, on average, the distributions only show minor drifts in comparisons of various

studied dimensions including the learned task, image domain, or dataset. However,

among the studied image domains, medical imaging models appeared to show

signiﬁcant outliers through “spikey” distributions, and, therefore, learn clusters of

highly speciﬁc ﬁlters different from other domains. Following this observation,

we study the collected medical imaging models in more detail. We show that

instead of fundamental differences, the outliers are due to speciﬁc processing in

some architectures. Quite the contrary, for standardized architectures, we ﬁnd that

models trained on medical data do not signiﬁcantly differ in their ﬁlter distributions

from similar architectures trained on data from other domains. Our conclusions

reinforce previous hypotheses stating that pre-training of imaging models can be

done with any kind of diverse image data.

1 Introduction

Deep learning has accelerated the progress in many computer vision problems, including various

applications in medical imaging [

]. Most deployed architectures are based on convolutional

neural networks (CNNs) which learn large amounts of convolution ﬁlters to transform inputs into

meaningful feature spaces that can be further processed. Since the same CNN architectures can

learn many computer vision tasks and work on diverse image domains, the question naturally arises

whether there are shared similarities in the learned convolution ﬁlters.

Previously, we introduced a large-scale dataset (

CNN Filter DB

) [

] that consists of convolution

ﬁlters with a kernel size of

3×3

extracted from hundreds of publicly available pre-trained CNNs

trained for a multitude of tasks. On that data, we performed a singular-value decomposition (SVD)

comparison of sets of ﬁlters aggregated by various dimensions of meta-data. Therefore, sets of

normalized ﬁlters were transformed to a common SVD-obtained basis, and the shift between resulting

distributions of the coefﬁcients (

cij

) was measured based on a variant of the Kullback-Leibler

Divergence (KL). Generally, individual models and model families (e.g.

ResNet

) learn unique

distributions of ﬁlters, but on average, distributions segregated by the learned task, image domain, or

dataset do not signiﬁcantly differ. Most observed shifts were due to different levels of “degenerated”

(i.e. sparse or highly repetitive) ﬁlters caused by the over-parameterization of networks relative to the

dataset. However, the most salient outlier in shifts included medical imaging models, as this domain

showed distinct “spikey” coefﬁcient distributions. In this paper, we set our focus to study these

models in detail and understand the possible causes of the shifts. The models can be summarized in

the following:

CompNet

[

] contains three customized architectures trained for brain segmentation

on the

OASIS

(MRI) dataset [

];

LungMask

[

] contains three

UNet

trained for lung segmentation on

the

LTRC

and some proprietary CT datasets;

TorchXRayVision

[

] contains seven

DensetNet-121

36th Conference on Neural Information Processing Systems (NeurIPS 2022).

arXiv:2210.13799v1 [eess.IV] 25 Oct 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

DoesMedicalImaginglearndifferentConvolutionFilters?PaulGavrikov1JanisKeuper1;21IMLA,OffenburgUniversity,Germany2FraunhoferITWM,Kaiserslautern,Germany{paul.gavrikov,janis.keuper}@hs-offenburg.deAbstractRecentworkhasinvestigatedthedistributionsoflearnedconvolutionltersthroughalarge-scalestudycontaini...

展开>> 收起<<

Does Medical Imaging learn different Convolution Filters Paul Gavrikov1Janis Keuper12.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Does Medical Imaging learn different Convolution Filters Paul Gavrikov1Janis Keuper12

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: