Automatic Assessment of Infant Face and Upper-Body Symmetry as Early Signs of Torticollis Michael Wan12 Xiaofei Huang2 Bethany Tunik3 Sarah Ostadabbas24

2025-05-02 0 0 6.86MB 6 页 10玖币

侵权投诉

Automatic Assessment of Infant Face and Upper-Body Symmetry

as Early Signs of Torticollis

Michael Wan1,2, Xiaofei Huang2, Bethany Tunik3, Sarah Ostadabbas2,4

1Roux Institute, Northeastern University, Portland, ME, USA

2Augmented Cognition Lab (ACLab), Northeastern University, Boston, MA, USA

3Board-Certiﬁed Clinical Specialist in Pediatric Physical Therapy, Boston, MA, USA

4Corresponding author: ostadabbas@ece.neu.edu

Abstract— We apply computer vision pose estimation tech-

niques developed expressly for the data-scarce infant domain to

the study of torticollis, a common condition in infants for which

early identiﬁcation and treatment is critical. Speciﬁcally, we use

a combination of facial landmark and body joint estimation

techniques designed for infants to estimate a range of geometric

measures pertaining to face and upper body symmetry, drawn

from an array of sources in the physical therapy and ophthal-

mology research literature in torticollis. We gauge performance

with a range of metrics and show that the estimates of most

these geometric measures are successful, yielding strong to

very strong Spearman’s ρcorrelation with ground truth values.

Furthermore, we show that these estimates, derived from pose

estimation neural networks designed for the infant domain,

cleanly outperform estimates derived from more widely known

networks designed for the adult domain1.

I. INTRODUCTION

Torticollis is a common condition in infants and children,

characterized by a persistent neck tilt or twist to one side. Its

most common form, congenital muscular torticollis (CMT),

has an estimated incidence of 3.9% to 16% [11]. Early treat-

ment is critical: outcomes are best when CMT is diagnosed

and physical therapy treatment started before the infant is

three months old, and conversely, if untreated or treated

later, CMT can lead to face, skull, or spine deformities,

pain and limited motion, and the need for invasive interven-

tions and surgery [17]. Screening, diagnosis, and monitoring

during treatment for CMT require laborious professional

assessments, so with the recent onset of computer vision

technology speciﬁcally studying infant face and body poses

for health and developmental applications [4], [8], [10], [13],

[20], it is natural to wonder whether algorithmic techniques

can help enable remote monitoring or automated screening

and diagnosis to augment clinical expertise. In this paper, as

a ﬁrst step towards such applications, we explore viability of

using computer vision techniques to assess a set of geometric

measures of symmetry in the face and upper body, previously

identiﬁed in the medical literature as being relevant to CMT

or the similarly presenting (non-congenital) ocular torticollis

condition, purely from casual photographs of infants in their

natural environments.

The geometric symmetry measures we consider are il-

lustrated in Fig. 1. We carefully researched and selected

1Code and data available at https://github.com/ostadabbas/Infant-Upper-

Fig. 1. We study the effectiveness of using deep learning computer vision

techniques to assess a range of geometric facial and upper body measures of

symmetry—illustrated schematically here—which are drawn from medical

research literature on torticollis in infants and children. The assessments

employ recent advances in infant-domain estimation of facial and upper

body landmarks—also illustrated faintly—and we demonstrate that this

yields better results than landmark estimation methods largely trained on

adult data.

these from among measures studied by physical therapy and

ophthalmology researchers to enable qualitative assessment

of changes over time and in response to treatments [1], [6],

[14], [22], including speciﬁcally to quantify outcomes after

surgery [3]. In [15], the authors even study the reliability

of the procedure of extracting such measurements from

still photographs itself. The general contention is not, of

course, that measurements from individual still photographs

are fully determinative of torticollis conditions, but rather

that aggregated over time they can be employed alongside

other tools as part of its detection and treatment. We hope

that by establishing the viability of algorithmic assessments

of these geometric measurements of symmetry from still

photographs, we will open the door to future applications in

automated monitoring and screening, including from videos.

The technical tools that we employ for this proof-of-

concept experiment are based in computer vision pose es-

timation from still images, and in particular, face and body

landmark estimation. Mature solutions to these tasks exist but

are generally based in deep learning from primarily adult

faces and bodies. As alluded earlier, specialized methods

tailored for the unique faces and bodies of infants have

arXiv:2210.15022v2 [eess.IV] 7 Nov 2022

only begun to crop up in recent years, in recognition of

the signiﬁcant domain gap between infant and adults from

the point of view of computer vision representations. In this

paper, we employ an infant face landmark estimation model

from [20] together with a infant body landmark estimation

model from [8], both of which employ domain adaptation

techniques to tune existing adult-focused models to the infant

domain. We work with a subset of the InfAnFace dataset

from [20], and compare the values of our six geometric

symmetry measures derived from both predicted and ground

truth landmarks. We propose modiﬁcations of these measures

from their original deﬁnitions in the literature to enable

compatibility with the landmarks used in pose estimation

techniques.

Our ﬁndings show that predictions derived from the recent

infant-domain pose models exhibit “strong” or “very strong”

Spearman’s ρranked correlation with the ground truth values

on a precisely labeled test set of infant faces in the wild, with

best performance on the gaze angle (ga) between the line

connecting the outer corners of the eyes and the midsternal

plumb line, and on the habitual head deviation (hhd), the

angle between the eye line and the acromion process (shoul-

der) line. Predictions of three other measurements (including

non-angles) were strong, but we found only moderate success

in the predictions of the orbit slopes angle (osa), the angle

between the lines connecting the outer and inner corners of

the eyes—this being arguably the most subtle metric. Based

on this and our further analysis involving more performance

metrics in Section V, we conclude that computer vision

infant pose estimation techniques can successfully measure

a range of quantities pertaining to torticollis.

II. BACKGROUND: TORTICOLLIS AND INFANT

DEVELOPMENT

A. Quantifying torticollis

While there is a large corpus of research and established

methodology in the diagnosis and treatment of torticollis, it

is largely based on in-person physical assessments by experts

and follow-ups with imaging or other deeper techniques [11].

By contrast, our work is inspired by a smaller cluster of

papers dealing explicitly with measuring signs and symptoms

of torticollis geometrically from still images.

We start with congenital muscular torticollis (CMT). The

author in [14] studied the effectiveness of a speciﬁc ther-

apeutic intervention for CMT by comparing changes in

an infant’s “habitual head deviation” (also “head tilt”) as

measured by hand from still photographs. The same author

later studied the reliability of this photograph-based method

itself, in [15]. Separately, [3] measured the “gaze angle”

and “transformational deformity” of child subjects, again

from still photographs, to gauge improvement in response

to surgical intervention. Measurements from photographs

offer researchers a repeatable, objective way to quantify the

change in severity of torticollis after an intervention.

Sometimes torticollis is not congenital (present from birth)

but rather acquired, as is the case with ocular torticollis,

where the abnormal head posture is adopted to compensate

for a defect in vision. In such cases, diagnoses often occur

only in adulthood, and can be informed by examination of

head pose in childhood photographs. Correspondingly, oph-

thalmologists have also studied the quantiﬁcation of facial

asymmetry via geometric measurements from still images.

An overview of such methods and quantities is given in [1],

who in turn cite [6] for deﬁnitions of facial measurements

such as the “orbit slopes angle,” “relative face size,” and

“facial bulk mass,” and [22] for deﬁnitions of the “facial

angle” and the “nasal tip deviation.” These measurements are

studied not as a means to quantify the effect of interventions,

but as part of a more comprehensive study on the differential

diagnosis of ocular torticollis and other conditions related to

facial asymmetries, especially superior oblique palsy [1].

B. Computer vision for infant health and development

We are not aware of prior computer vision research

intended to detect torticollis or gauge head asymmetry in

infants. The closest in spirit might be a pair of related papers,

[19], [24], in which researchers employ computer vision

to analyze head posture and tremor with a view towards

algorithmic understanding of cervical dystonia (also known

as spasmodic torticollis), with incidence largely in the adult

population. Accordingly, those studies can take advantage

of far more mature adult-focused head pose estimation

techniques like OpenFace 2.0 [2], whereas our efforts are

highly constrained by data scarcity in the infant domain. In

the infant domain, there is prior work on bodies but not

faces: [4] develops an infant-speciﬁc body pose estimation

deep network to extract body motion information from infant

videos, in an attempt to assess infant neuromotor risk; and

[10] uses 3D infant pose estimation techniques to assess

infant body symmetry, with a view towards applications in

infant development and torticollis, but without a speciﬁc im-

plementation of such. As mentioned, all of this work comes

in the context of recent attention in computer vision to the

small data domain problem of infant pose comprehension,

for both faces [20] and bodies [7], [8], [9], [13].

III. CONCEPTS: MEASUREMENTS OF

SYMMETRY

From the sources cited in Section II-A, we identiﬁed all

clearly deﬁned geometric symmetry measures used by re-

searchers in the study of torticollis and facial asymmetries—

six in all. We altered the deﬁnitions to base them explicitly

on the 68 facial landmark coordinates and two body joint

(shoulder) coordinates used by our pose estimators, as illus-

trated and enumerated in Fig. 2, which also enables more

consistent comparisons. The ﬁnal measures are deﬁned in

Table I and illustrated in Fig. 1. In the rest of this section,

we clarify and discuss these deﬁnitions.

A. Assumptions and context

Since we work with both ground truth and estimated land-

mark coordinates in two dimensions, we generally assume

that every infant is facing forward into the camera, so that the

infant’s face plane is roughly parallel with the camera image

plane. In principle, all of the measurements we consider are

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：

AutomaticAssessmentofInfantFaceandUpper-BodySymmetryasEarlySignsofTorticollisMichaelWan1;2,XiaofeiHuang2,BethanyTunik3,SarahOstadabbas2;41RouxInstitute,NortheasternUniversity,Portland,ME,USA2AugmentedCognitionLab(ACLab),NortheasternUniversity,Boston,MA,USA3Board-CertiedClinicalSpecialistinPediatric...

展开>> 收起<<

Automatic Assessment of Infant Face and Upper-Body Symmetry as Early Signs of Torticollis Michael Wan12 Xiaofei Huang2 Bethany Tunik3 Sarah Ostadabbas24.pdf

共6页,预览2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！

Automatic Assessment of Infant Face and Upper-Body Symmetry as Early Signs of Torticollis Michael Wan12 Xiaofei Huang2 Bethany Tunik3 Sarah Ostadabbas24

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: