landmark localization and registration of 3d facial scans for the evaluation of orthodontic...

Landmark localization and registration of 3D facial scans

for the evaluation of orthodontic treatments in maxillofacial and oral surgery

School of EECS: Prathap Nair, Dr Andrea Cavallaro

School of Medicine and Dentistry: Dr Lifong Zou

Mid-project update

What is the problem?

• To quantify 3D facial asymmetry

• Clinical diagnosis• Treatment planning• Post-treatment monitoring• Statistical studies on a large population

• What is rigid registration?• Alignment of 2 or more faces

• Classical approach: Iterative Closest Point (ICP) algorithm• Advantage

• no prior info needed• Disadvantage

• random points used for matching can lead to erroneous results

Approach: rigid registration

Example

Our approach

• Rigid registration based on landmarks • Landmark detection via Statistical Shape Analysis

BtG project: Achievement 1

• Improved accuracy

Red – before BtG

Green – after BtG

Approach: overview

Test Scan

Reference scan

Detection of

Landmark

Points

Detection of

Landmark

Points

Coarse registration using

Key landmark

s

Detection of Stable regions

Fine registrat

ion using the

Semantic

Regions

Distance

estimation

Test scan Reference scan

Approach: overview

Test Scan

Reference scan

Detection of

Landmark

Points

Detection of

Landmark

Points

Coarse registration using

Key landmark

s

Detection of Stable regions

Fine registrat

ion using the

Semantic

Regions

Distance

estimation

Test scan Reference scan

Approach: overview

Test Scan

Reference scan

Detection of

Landmark

Points

Detection of

Landmark

Points

Coarse registration using

Key landmark

s

Detection of Stable regions

Fine registrat

ion using the

Semantic

Regions

Distance

estimation

Test scan Reference scan

Approach: overview

Test Scan

Reference scan

Detection of

Landmark

Points

Detection of

Landmark

Points

Coarse registration using

Key landmark

s

Detection of Stable regions

Fine registrat

ion using the

Semantic

Regions

Distance

estimation

Key Landmar

ks

Coarse registration

Test scan Reference scan

Approach: overview

Test Scan

Reference scan

Detection of

Landmark

Points

Detection of

Landmark

Points

Coarse registration using

Key landmark

s

Detection of Stable regions

Fine registrat

ion using the

Semantic

Regions

Distance

estimation

Test scan Reference scan

Approach: overview

Test Scan

Reference scan

Detection of

Landmark

Points

Detection of

Landmark

Points

Coarse registration using

Key landmark

s

Detection of Stable regions

Fine registrat

ion using the

Semantic

Regions

Distance

estimation

Fine registration

Approach: overview

Test Scan

Reference scan

Detection of

Landmark

Points

Detection of

Landmark

Points

Coarse registration using

Key landmark

s

Detection of Stable regions

Fine registrat

ion using the

Semantic

Regions

Distance

estimation

ICP Proposed approach

Example

BtG project: Achievement 2

• User friendly GUI• To ease burden on clinicians• User-feedback mechanisms

Conclusions

• Achievements• Improved landmark localisation accuracy • More user-friendly GUI with the user feedback

• Current work• Clinical evaluation of the landmark detection accuracy • Validation of 3D facial scan registration accuracy• Further improving the GUI based on clinician feedback

Contact:

prathap.nair@elec.qmul.ac.uk

andrea.cavallaro@elec.qmul.ac.uk