3D Riesz-wavelet Based Covariance Descriptors for Texture Classication of Lung Nodule Tissue in CT

IntroductionMethodology

Experimental evaluationConclusions

3D Riesz–wavelet Based Covariance Descriptorsfor Texture Classification ofLung Nodule Tissue in CT

Pol Cirujeda./, Henning Muller†, Daniel Rubin?,Todd A. Aguilera?, Billy W. Loo Jr.?,

Maximilian Diehn?, Xavier Binefa./, Adrien Depeursinge†,‡

./ † ? ‡

SaAT6.6

August 29th, 20151 / 18



Motivation / ContributionRelated Work

Motivation + Contribution

I (Statistical) feature extraction and representation

I Dictionary modelling of lung/nodule tissue areas

I Classification of lung nodule areas(solid / ground glass opacity -GGO / healthy)

Goal and application

Supervised learning of region classes from textureRobustness to size and shape variations

Applications: tissue modelling, classification, segmentation

2 / 18




Motivation + Contribution

I (Statistical) feature extraction and representation

I Dictionary modelling of lung/nodule tissue areas

I Classification of lung nodule areas(solid / ground glass opacity -GGO / healthy)

Goal and application

Supervised learning of region classes from textureRobustness to size and shape variations

Applications: tissue modelling, classification, segmentation

3 / 18




Related work

Clinical domain

I 3D visualization and annotation software

I Manual delineation with expertise of clinicians

Computer vision + Machine Learning domain

I 3D features: Riesz transform, 1st and 2nd order visual cues

I 3D descriptors: MCOV, 3D-SIFT, SHOT, THRIFT...

I Linear/non-linear (un)supervised classification methods:CNN, Kernel-SVMs, Sparse coding, Bag-of-visual features...

4 / 18




Related work

Clinical domain

I 3D visualization and annotation software

I Manual delineation with expertise of clinicians

Computer vision + Machine Learning domain

I 3D features: Riesz transform, 1st and 2nd order visual cues

I 3D descriptors: MCOV, 3D-SIFT, SHOT, THRIFT...

I Linear/non-linear (un)supervised classification methods:CNN, Kernel-SVMs, Sparse coding, Bag-of-visual features...

5 / 18



IntuitionRiesz-Covariance descriptorsData and patientsClassification

Intuition - features

Riesz-wavelet transform as texture features 1

Figure 1: Lung nodule CT slice with corresponding Riesz filter responses

1A. Depeursinge et al., ”Lung Texture Classification Using Locally–Oriented RieszComponents”, in MICCAI 2011

6 / 18




Intuition - feature representation (in 3D)

3D Riesz-covariance models

Figure 2: 3D CT volume and associated Riesz-covariance descriptor

7 / 18




3D Riesz-Covariance descriptors

x y

zv

Φ(ct, v) = {φx ,y ,z , ∀x , y , z ∈ v} (1)

φx ,y ,z =(R(n1,n2,n3)

x ,y ,z , ‖R‖x ,y ,z , ctx ,y ,z)

(2)

RieszCov (Φ(ct, v)) =1

N − 1

N∑i=1

(φx ,y ,z − µφ)) (φx ,y ,z − µφ))T ,

(3)

8 / 18




Covariance model benefits

I Common framework for statistical data modelling.

I Features ≡ samples of n−dim joint distributions.

I Second order moment statistics (n × n covariance matrices).

I Covariances manifold (Sym+d ) ⇒ analytical modelling

Riemannian space ported machine learning techniques.

9 / 18




Covariance descriptors - Sym+d Riemannian space

logId

TId

x = logY (X ) = Y12 log

(Y−

12XY−

12

)Y

12 (4)

x = vect(x) = (x1,1, x1,2, ..., x1,d , x2,2, x2,3, ..., xd ,d) (5)

δ(X ,Y ) =

√Trace

(log(X−

12YX−

12

))(6)

10 / 18




Data gathering

Ground-truth:

I 95 patients (from Stanford Hospital and Clinics)

I Biopsy-proven early stage non-small cell lung carcinoma

I Nodule regions delineated in CTs by clinicians

I Processing with MATLAB software:isotropic voxels of 0.8 mm3

11 / 18




Data samples

GGO vs. Solid lung nodule tissue components (vs. healthy lung)

Figure 3: Lung nodule Riesz filterresponses - GGO component

Figure 4: Lung nodule Riesz filterresponses - solid component

12 / 18




Bag-of-covariances

Standard bag-of-visual features paradigm.Sub-sampling of partial class regions for a complete dictionarymodelling

13 / 18




Bag-of-covariances

I Dictionary D ≡ x cv ,p = vect(logId (RieszCovCV ,P))

I Training set: modelling frequencies of words in D

I Classification decision: class(ct) = argmini D(hct , hi )

14 / 18



Experimental setup

Experimental setup

Data sets:

I 35 patients for model learning, 60 for test set.

I 60 words per class (dictionary size, 3× 60× 35 = 3600)

I 10-fold cross-validation.

Quantitative evaluation w.r.t. ground-truth:

I Avg. sensitivity (TP / TP+FN) = 82.2%

I Avg. specificity (TN / TN+FP) = 86.2%

15 / 18



RemarksFuture work

Conclusions

I Computer vision and Machine Learning to the service ofmedical knowledge

I Statistical design for a robust solution

I Easily extendible framework

16 / 18



RemarksFuture work

Future work

I More patients and bigger data collection

I Different modelling contexts for concrete problems

I Exploit the covariance-based descriptor space for differentMachine Learning techniques (clustering, classification,regression...)

17 / 18



RemarksFuture work

Thanks for your attention

Questions?

3D Riesz–wavelet Based Covariance Descriptorsfor Texture Classification ofLung Nodule Tissue in CT

SaAT6.6

Pol Cirujeda, UPF

EMBC 2015, August 29th 2015

18 / 18

3D Riesz-wavelet Based Covariance Descriptors for Texture Classication of Lung Nodule Tissue in CT

Technology

Transcript of 3D Riesz-wavelet Based Covariance Descriptors for Texture Classication of Lung Nodule Tissue in CT