Semi-Automatic Central-Chest Semi-Automatic Central-Chest Lymph-Node Definition from Lymph-Node Definition from

3D MDCT Images3D MDCT ImagesKongkuo Lu and William E. Higgins

Penn State University

University Park and Hershey, PA, USA

SPIE Medical Imaging 2010: Computer-Aided Diagnosis, San Diego, CA, Feb. 17, 2010

Drawing by Terese Winslow, “Bronchoscopy,” Drawing by Terese Winslow, “Bronchoscopy,” NCI Visuals OnlineNCI Visuals Online, National , National Cancer InstituteCancer Institute

MotivationMotivation Lung cancer is the Lung cancer is the

leading cause of cancer leading cause of cancer deathdeath

Early lung cancer Early lung cancer detection could increase detection could increase survival ratesurvival rate

Diagnosis procedures:Diagnosis procedures: 3D MDCT chest 3D MDCT chest

image assessmentimage assessment Follow-on diagnostic Follow-on diagnostic

bronchoscopybronchoscopy

512 x 512 x ~700 voxels512 x 512 x ~700 voxels-1000 HU to 1000 HU -1000 HU to 1000 HU

(grayscale)(grayscale)

MotivationMotivation

Armato Armato et al.et al., , RadiologyRadiology 2004 2004 Subaortic and subcarinal lymph node (p2h012b)Subaortic and subcarinal lymph node (p2h012b)

Lower para-tracheal lymph node (Lower para-tracheal lymph node (IRB20349.3.3IRB20349.3.3))

ROIs involve complex ROIs involve complex phenomenaphenomena

Manual slice tracing – not feasibleManual slice tracing – not feasible Automatic methods Automatic methods

– – application dependentapplication dependent Semi-automatic methods Semi-automatic methods

– – more practicalmore practicalLive WireLive Wire

Prior WorkPrior Work

Mortensen and Barrett (Mortensen and Barrett (Graphical Models and Medical Imaging Graphical Models and Medical Imaging 19981998))

FalcFalcão ão et al. et al. ((Graphical Models and Medical Imaging 1998Graphical Models and Medical Imaging 1998)) Lu and HigginsLu and Higgins ((Int. J. CARS 2007)Int. J. CARS 2007)

2D Live Wire:2D Live Wire:

Falcão and Udupa (Medical Image Analysis 2000) Hamarneh et al. (SPIE Medical Imaging 2005) König and Hesser (SPIE Medical Imaging 2005) Souza et al. (SPIE Medical Imaging 2006) Lu and HigginsLu and Higgins ((Int. J. CARS 2007)Int. J. CARS 2007) Poon et al. (SPIE Medical Imaging 2007 and CMIG 2008)

3D Live Wire:3D Live Wire:

Prior WorkPrior Work

),(),()()(),( qpfwqpfwqfwqfwqpl DfDfDDGGzz Cost Function:Cost Function:

292924242222181815151414448810101919

404028282020141413131111551100551111

3333222219192020131388448833991616

3030242427273636292924241414121277771515

3434292937376262393933331919131310101414

44442929161613132020

29291919171719192929

25252323252529294747

3333303035354141

Dynamic Graphic Search:Dynamic Graphic Search:

Mortenson Mortenson GMMIGMMI 1998; K. Lu and W.E. Higgins, 1998; K. Lu and W.E. Higgins, SPIESPIE 2006 and 2006 and IJCARIJCAR 2007 2007

Prior WorkPrior Work

K. Lu and W.E. Higgins, K. Lu and W.E. Higgins, SPIESPIE 2006 and 2006 and IJCARIJCAR 2007 2007

Segment ROI using 2D Live WireSegment ROI using 2D Live WireSegment ROI using 3D Live WireSegment ROI using 3D Live Wire

Central-Chest Lymph Nodes in 3D MDCT Image

Typical lower paratracheal (M4) Typical lower paratracheal (M4) lymph nodeslymph nodes

M4-3M4-3

M4-1M4-1

M4-1M4-1

M4-2M4-2

CoroCoro

TransTrans

Complete segmentation of Complete segmentation of

central-chest lymph nodescentral-chest lymph nodes

Case IRB20349.3.3Case IRB20349.3.3

Single-Section Live Wire

Single-Section Live WireAdjust working area adaptivelyAdjust working area adaptively

Refine seed set iterativelyRefine seed set iteratively

Terminate iterations whenTerminate iterations when

Terminate 3D process whenTerminate 3D process when

Section limits reachedSection limits reached

Boundary costs vary greatlyBoundary costs vary greatly

Boundary is too smallBoundary is too small

Segmented region is too Segmented region is too

smallsmall

Intensity distribution varies Intensity distribution varies

greatlygreatlyK. Lu, PhD Dissertation 2010K. Lu, PhD Dissertation 2010

Single-Section Live Wire

Single-Section Live Wire

Single-Click Live Wire

Single-Click Live Wire

Single-Section and Single-Click LW

Single-SectionSingle-Section Single-ClickSingle-Click

Lu, PhD Dissertation 2010 and K. Lu and W.E. Higgins Lu, PhD Dissertation 2010 and K. Lu and W.E. Higgins Computer in Biology and Medicine Computer in Biology and Medicine (in submission)(in submission)

Performance EvaluationPerformance Evaluation

Computer - Dell Precision 650 workstation:Computer - Dell Precision 650 workstation:

MDCT Data:MDCT Data:

• Dual Intel Xeon 3.2GHz, 3GB RAM, Windows XPDual Intel Xeon 3.2GHz, 3GB RAM, Windows XP

ResultsResults

Segmented Nodes: Short-Axis Length: 5.8±1.5 mm±1.5 mm Long-Axis Length:Long-Axis Length: 11±4.0 mm 11±4.0 mm Volume: 256±210 mm±210 mm33

Number of Voxels: 1089±861±861

Method Single-Section Single-Click

Number of Nodes 50

Success Rate 90 % (45/50) 80% (40/50)

Accuracy 81±7% 79±8%

Inter-Trial Reproducibility 88±7% 86±9%

Processing Time 16±4s 20±5s

Comparison of Two ObserversComparison of Two Observers

Operator IndependentOperator Independent

20349.3.15 21405.67

O1 O2 O1 O2

Accuracy 83% 80% 78% 79%

Reproducibility 87% 88% 89% 85%

Processing Time 16s 20s 19s 22s

Example Lymph Node SegmentationsExample Lymph Node Segmentations

21405.6421405.64 20349.3.320349.3.3

* Segmentable nodes: both observers and both methods * Segmentable nodes: both observers and both methods

SummarySummary

Single-Section and Single-Click Live Wire

Reduce human interaction

Efficient and reliable

Operator independent

Can handle typical lymph nodes and other ROIs

• 90% Single-Section

• 80% Single-Click

AcknowledgmentsAcknowledgments

• NIH NCI grant #R01-CA074325NIH NCI grant #R01-CA074325• Pinyo Taeprasartsit - experiment participantPinyo Taeprasartsit - experiment participant

The Multidimensional Imaging Processing Lab at The Multidimensional Imaging Processing Lab at Penn StatePenn State

Single-Section and Single-Click Single-Section and Single-Click LW EvaluationLW EvaluationLymph node #: 50Succ. Seg. : M1-45 (90%) M2-40 (80%)Accuracy: >80%Inter-trial Reprod.: >85%

MM11, M, M22:Single-section and single-click LW; :Single-section and single-click LW; εε11, , εε22: Trials; a(: Trials; a(MMii, , εεjj): accuracy; r(): accuracy; r(MMii, , εεjj, , εεkk): reproducibility): reproducibility

Single-Section and Single-Click Single-Section and Single-Click LW EvaluationLW Evaluation

MM11, M, M22:Single-section and single-click LW; :Single-section and single-click LW; εε11, , εε22: Trials; a(: Trials; a(MMii, , εεjj): accuracy; r(): accuracy; r(MMii, , εεjj, , εεkk): reproducibility): reproducibility