A STRONGER BASELINE FOR AUTOMATIC PFIRRMANN GRADING OF LUMBAR SPINE MRI USING DEEP LEARNING Narasimharao Kowlagi Huy Hoang Nguyen Terence McSweeney

2025-04-30 0 0 895KB 5 页 10玖币

侵权投诉

A STRONGER BASELINE FOR AUTOMATIC PFIRRMANN GRADING OF LUMBAR SPINE

MRI USING DEEP LEARNING

Narasimharao Kowlagi Huy Hoang Nguyen Terence McSweeney

Simo Saarakkala Juhani M¨

a¨

att¨

a Jaro Karppinen Aleksei Tiulpin

University of Oulu

ABSTRACT

This paper addresses the challenge of grading visual features

in lumbar spine MRI using Deep Learning. Such a method is

essential for the automatic quantiﬁcation of structural changes

in the spine, which is valuable for understanding low back

pain. Multiple recent studies investigated different architec-

ture designs, and the most recent success has been attributed

to the use of transformer architectures. In this work, we ar-

gue that with a well-tuned three-stage pipeline comprising se-

mantic segmentation, localization, and classiﬁcation, convo-

lutional networks outperform the state-of-the-art approaches.

We conducted an ablation study of the existing methods in

a population cohort, and report performance generalization

across various subgroups. Our code is publicly available to

advance research on disc degeneration and low back pain.

Index Terms—Lumbar Spine, Pﬁrrmann Grading, Disc

Degeneration, Deep Learning, Convolutional Neural Network

1. INTRODUCTION

Lumbar Disc Degeneration (LDD) was found to be one

of several reasons causing Low Back Pain [1]. Although

there are several grading systems to quantify LDD, the Pﬁr-

rmann and Schneiderman scales are commonly used [2].

Pﬁrrmann et al. proposed a 5-scale grading system using

T2-weighted MR imaging [3]. The grading system uses an

algorithmic approach for LDD using signal intensity and disc

homogeneity, demonstrated in Fig. 1. Over the past few years,

several Deep Learning based methods were built to quantify

LDD using MR imaging and the Pﬁrrmann grading system.

In a study by da Silva Barreiro et al. [4], the interver-

tebral discs (IVDs) were extracted using manually segmented

binary masks, and an artiﬁcial neural network was used to per-

form Pﬁrrmann grading (PG). Later, SpineNet [5] – a frame-

work that can automatically grade spine MRIs for different

spine disorders, including LDD using the PG system was pro-

posed. The authors used a graph-based approach to detect

and group the vertebral bodies (VB) and discs in their work.

Corresponding Author: narasimharao.kowlagi@oulu.ﬁ

(a) PG 2 (b) PG 3 (c) PG 4 (d) PG 5

Fig. 1: Pﬁrrmann grades in L5-S1 as graded in the Northern

Finland Birth Cohort dataset

Then, a VGG-M based architecture was used as the classi-

ﬁer. Newer architectures and methods were subsequently ex-

ploited in other works. For instance, SpineNetV2 [6] used

a UNet [7] based architecture to improve vertebrae detection,

and a ResNet [8] architecture was used for the classiﬁcation of

the spine disorders. Although the improvement was not sub-

stantial, this work highlighted the impact of design choices on

the overall outcome.

More recently, a transformer [9] based architecture for

spine grading [10] was proposed and it outperformed the ar-

chitecture designs proposed in the earlier works. In another

related work, DeepSpine [11] used a ResNeXt50 [12] archi-

tecture with the UNet model for spine segmentation to detect

stenosis. Additionally, it is noteworthy to see an external val-

idation of these methods [13].

Although there is continuous improvement to create bet-

ter spine grading models with newer architectures, we ﬁnd a

lack of thorough empirical evaluation in the existing research.

Furthermore, current studies leveraged clinical data, but in re-

ality, population cohorts are used in research. In this study, we

take a step towards addressing these two challenges. Speciﬁ-

cally, the contributions of our work are:

1. We revisit the design choices used to create architectures

for lumbar spine MRI grading, and propose a strong con-

volutional neural network (CNN) based method that out-

performs the existing approaches in a systematic evalua-

tion.

2. Compare to the existing methods, we evaluate our method-

ology on a population cohort, which reﬂects the transfer-

ability of our results to downstream research applications.

3. Finally, we make the code of this paper and the pre-trained

arXiv:2210.14597v1 [eess.IV] 26 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：

ASTRONGERBASELINEFORAUTOMATICPFIRRMANNGRADINGOFLUMBARSPINEMRIUSINGDEEPLEARNINGNarasimharaoKowlagiHuyHoangNguyenTerenceMcSweeneySimoSaarakkalaJuhaniM¨a¨att¨aJaroKarppinenAlekseiTiulpinUniversityofOuluABSTRACTThispaperaddressesthechallengeofgradingvisualfeaturesinlumbarspineMRIusingDeepLearning.Sucham...

收起<<

A STRONGER BASELINE FOR AUTOMATIC PFIRRMANN GRADING OF LUMBAR SPINE MRI USING DEEP LEARNING Narasimharao Kowlagi Huy Hoang Nguyen Terence McSweeney.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！

A STRONGER BASELINE FOR AUTOMATIC PFIRRMANN GRADING OF LUMBAR SPINE MRI USING DEEP LEARNING Narasimharao Kowlagi Huy Hoang Nguyen Terence McSweeney

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: