SPATIO-TEMPORAL HYBRID FUSION OF CAE AND SWIN TRANSFORMERS FOR LUNG CANCER MALIGNANCY PREDICTION Sadaf Khademiy Shahin Heidarianz Parnian Afshary Farnoosh NaderkhaniyAnastasia Oikonomouyy

2025-05-03 0 0 851.28KB 5 页 10玖币

侵权投诉

SPATIO-TEMPORAL HYBRID FUSION OF CAE AND SWIN TRANSFORMERS FOR LUNG

CANCER MALIGNANCY PREDICTION

Sadaf Khademi†, Shahin Heidarian‡, Parnian Afshar†, Farnoosh Naderkhani†Anastasia Oikonomou††,

Konstantinos N. Plataniotis‡‡, and Arash Mohammadi†

†Concordia Institute for Information Systems Engineering, Concordia University, Montreal, Canada

‡Department of Electrical and Computer Engineering, Concordia University, Montreal, Canada

††Department of Medical Imaging, Sunnybrook Health Sciences Centre, Toronto, Canada

‡‡Department of Electrical and Computer Engineering, University of Toronto, Toronto, Canada

ABSTRACT

The paper proposes a novel hybrid discovery Radiomics framework

that simultaneously integrates temporal and spatial features extracted

from non-thin chest Computed Tomography (CT) slices to predict

Lung Adenocarcinoma (LUAC) malignancy with minimum expert

involvement. Lung cancer is the leading cause of mortality from can-

cer worldwide and has various histologic types, among which LUAC

has recently been the most prevalent. LUACs are classiﬁed as pre-

invasive, minimally invasive, and invasive adenocarcinomas. Timely

and accurate knowledge of the lung nodules malignancy leads to a

proper treatment plan and reduces the risk of unnecessary or late

surgeries. Currently, chest CT scan is the primary imaging modality

to assess and predict the invasiveness of LUACs. However, the radi-

ologists’ analysis based on CT images is subjective and suffers from

a low accuracy compared to the ground truth pathological reviews

provided after surgical resections. The proposed hybrid framework,

referred to as the CAET-SWin, consists of two parallel paths: (i) The

Convolutional Auto-Encoder (CAE) Transformer path that extracts

and captures informative features related to inter-slice relations via

a modiﬁed Transformer architecture, and; (ii) The Shifted Window

(SWin) Transformer path, which is a hierarchical vision transformer

that extracts nodules’ related spatial features from a volumetric CT

scan. Extracted temporal (from the CAET-path) and spatial (from the

Swin path) are then fused through a fusion path to classify LUACs.

Experimental results on our in-house dataset of 114 pathologically

proven Sub-Solid Nodules (SSNs) demonstrate that the CAET-SWin

signiﬁcantly improves reliability of the invasiveness prediction task

while achieving an accuracy of 82.65%, sensitivity of 83.66%, and

speciﬁcity of 81.66% using 10-fold cross-validation.

Index Terms—Lung Adenocarcinoma, Lung Nodule Invasive-

ness, Transformer, Subsolid Nodule, Self-Attention.

1. INTRODUCTION

Lung Cancer (LC) is the deadliest and least-funded cancer world-

wide [1, 2]. Non-small-cell LC is the major type of LC, and Lung

Adenocarcinoma (LUAC) is the most prevalent histologic sub-

type [3]. Lung nodules manifesting as Ground Glass (GG) or Sub-

solid Nodules (SSNs) on Computed Tomography (CT) scans have

a higher risk of malignancy than other incidentally detected small

solid nodules. SSNs are often diagnosed as adenocarcinoma and

are generally classiﬁed into pure GG nodules and part-solid nodules

according to their appearance on the lung window settings [4, 5].

A timely and accurate attempt to differentiate the LUACs is of

utmost importance to guide a proper treatment plan, as in some

cases, a pre-invasive or minimally invasive SSN can be monitored

with regular follow-up CT scans, whereas invasive lesions should

undergo immediate surgical resection if they are deemed eligible.

Most often, the SSNs type is diagnosed based on the pathological

ﬁndings performed after surgical resections, which is not desired

for prior treatment planning. Currently, radiologists use chest CT

scans to assess the invasiveness of the SSNs based on their imag-

ing ﬁndings and patterns prior to making decisions regarding the

appropriate treatment. Such visual approaches, however, are time-

consuming, subjective, and error-prone. So far, many studies have

used high-resolution and thin-slice (<1.5mm) CT images for the

SSN classiﬁcation, which require longer analysis times, as well as

more reconstruction time [6, 7]. However, lung nodules are mostly

identiﬁed from CT scans performed for varied clinical purposes

acquired using routine standard or low-dose scanning protocols with

non-thin slice thicknesses (up to 5mm) [8]. In addition, recent

lung cancer screening recommendation, suggests using low-dose

CT scans with thicker slice-thicknesses (up to 2.5mm) [9, 10].

Capitalizing on the above discussion, the necessity of developing

an automated invasiveness assessment framework that performs

well regardless of technical settings has recently arisen among the

research community and healthcare professionals.

Related Works: Generally speaking, existing works on the SSN in-

vasiveness assessment can be categorized into two main classes: (i)

Radiomics-based, and; (ii) Deep Learning-based frameworks, also

referred to as Discovery Radiomics [11]. In the former class, data-

characterization algorithms extract quantitative features from nodule

masks and the original CT images, which are then analyzed using

statistical or conventional Machine Learning (ML) models [12, 13].

As an example of such frameworks, a histogram-based model is de-

veloped in [8] to predict the invasiveness of primary adenocarcinoma

SSNs from non-thin CT scans of 109 pathologically labeled SSNs.

In this study, a set of histogram-based and morphological features

along with additional features extracted via the functional Princi-

pal Component Analysis (PCA) is fed to a linear logistic regression.

Discovery Radiomics approaches, on the other hand, extract infor-

mative and discriminative features in an automated fashion. Existing

deep models working with volumetric CT scans can be classiﬁed in

two categories: (i) 3D-based solutions [14], where the whole 3D vol-

ume of CT images are fed to the model. Processing a large 3D CT

scan at once, however, results in extensive computational complex-

ity requiring more computational resources, and enormous training

datasets, and; (ii) 2D-based solutions [15–17], where individual 2D

CT slices are ﬁrst analyzed, which are then fused via an aggregation

arXiv:2210.15297v1 [eess.IV] 27 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：

SPATIO-TEMPORALHYBRIDFUSIONOFCAEANDSWINTRANSFORMERSFORLUNGCANCERMALIGNANCYPREDICTIONSadafKhademiy,ShahinHeidarianz,ParnianAfshary,FarnooshNaderkhaniyAnastasiaOikonomouyy,KonstantinosN.Plataniotiszz,andArashMohammadiyyConcordiaInstituteforInformationSystemsEngineering,ConcordiaUniversity,Montreal,Can...

展开>> 收起<<

SPATIO-TEMPORAL HYBRID FUSION OF CAE AND SWIN TRANSFORMERS FOR LUNG CANCER MALIGNANCY PREDICTION Sadaf Khademiy Shahin Heidarianz Parnian Afshary Farnoosh NaderkhaniyAnastasia Oikonomouyy.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！

SPATIO-TEMPORAL HYBRID FUSION OF CAE AND SWIN TRANSFORMERS FOR LUNG CANCER MALIGNANCY PREDICTION Sadaf Khademiy Shahin Heidarianz Parnian Afshary Farnoosh NaderkhaniyAnastasia Oikonomouyy

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: