INTRODUCTION

Estimation of Adipose Compartment Volumes in CT Images of a Mastectomy Specimen

Abdullah-Al-Zubaer Imran*, David D. Pokrajac*, Andrew D. A. Maidment**, Predrag R. Bakic**

*Delaware State University, **University of Pennsylvania

CT ACQUISITION & RECONSTRUCTION

MASTECTOMY SPECIMEN CT SLICES

ADIPOSE COMPARTMENT SEGMENTATION

RESULT: COMPARTMENTAL VOLUME

RESULTS: DISTRIBUTION OF VOLUME

RESULTS: CONFIDENCE VS. VOLUME

RESULTS: TIME VS. VOLUME

PURPOSE

WORK IN PROGRESS

CONCLUSIONS

REFERENCES

ACKNOWLEDGEMENT

[1] Pokrajac, D. D., Maidment, A. D. A. and Bakic, P. R., “Optimized generation of high resolution breast anthropomorphic software phantoms,” Med. Phys. 39(4), 2290-2302 (2012).

[2] Yushkevich, P. A., Piven, J., Hazlett, H. C., Smith, R. G., Ho, S., Gee, J.C., and Gerig, G., “User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability,” Neuroimage 31(3), 1116-28 (2006).

[3] Imran, A. -A. -Z., Bakic, P. R. and Pokrajac, D. D., “Spatial distribution of adipose compartments size, shape and orientation in a CT breast image of a mastectomy specimen,” IEEE SPMB Symp Dec. 13 (2015).

CT slices were imported in

ITK-SNAP software [2].

306 slices were analyzed;

remaining slices did contain

no tissue or had poor quality.

View of CT slices in sagittal plane

Curve-based contrast adjustment was

performed and slices were viewed in

sequence for better understanding of

the compartments in consecutive slices. Axial plane view of slices

Slice views in coronal plane

The compartments were identified and then segmented

from each slice manually with the boundary marking.

205 most discernible compartments were segmented

from 619 input slices.

Segmentation marked slices in axial, sagittal and coronal planes

3-D visualizations of segmented compartments

The segmented compartments spanned 4082

slices, approx. 20 on average.

The average estimated volume was

0.91 cm3 ± 0.87.

Scatter plot of segmentation time Vs. volume

Distribution of volume was not normal;

rather left-skewed.

Kolmogorov-Smirnov test, Lilliefors test,

Jarque-Bera test, and visual cdf

comparison (between empirical and

standard) rejected the hypothesis of

the normality of volume data.

Per slice segmentation confidence level

was assigned in the scale of 5 (1-5).

The average confidence for 205

segmented compartments was 3.88. Compartment volume was not

correlated with confidence.

The confidence level was assigned

based upon operator’s visibility rather

than the size of a compartment.

Histogram of average confidence level (1-5)

Confidence level Vs. estimated volume

Average time spent for segmenting a compartment

was 8.75 minutes.

The estimated volume was correlated to the

segmentation time (p < 0.001).

Standard deviation of residuals tends to increase with

volume indicating heteroscedasticity.

Selection bias and small path sample may have caused high variance in volume data.

Anatomical measurements of the breast tissue size and distribution of

adipose compartments from High Resolution CT slices of a mastectomy

specimen.

Virtual clinical trials (VCTs) based upon the computer simulation of

breast anatomy, imaging, and image analysis, represent a viable

preclinical alternative to the conventional clinical trials. Realistic

simulation urges for the measurements of breast anthropometrics from

real clinical breast images [1].

A total mastectomy specimen was imaged on a whole body, multi-slice CT

system (Siemens Sensation 64) using the following parameters:

Shape analysis and distribution of segmented adipose

compartments size, shape, and orientation [3].

Identification and characterization of the adipose

compartments automatically in clinical CT or MRI

images.

This work is the proof of the concept for manual

volumetric estimation of adipose compartments in

breast. Automatizing the segmentation would

provide faster and more decisive measurements.

Extracted measures could be utilized in the

simulation for the development of more realistic

software breast phantoms for VCT.

Each compartment spanned in

multiple slices.

There were some compartments,

clearly distinguishable from neighbors.

Adipose Compartments

Histogram of #slices in adipose compartments

Volumes of the segmented adipose compartments (cm3)

Segm

enta

tio

n t

ime

for

com

par

tmen

ts (

sec)

Volumes of the segmented adipose compartments (cm3)

Avg

. co

nfi

den

ce le

vel o

f co

mp

artm

ents

(1

-5)

Standard normal quantiles

Qu

anti

les

of

volu

me

inp

ut

sam

ple

QQ plot of sample volume data and

standard normal

Histogram of estimated compartment volumes (cm3)

This research was supported by a grant from the National Institute of General Medical Sciences (P20 GM103446) from the National Institutes of Health. Also, the work was supported in part by the US Department of Defense Breast Cancer Research Program (HBCU Partnership Training Award #BC083639), the US National Science Foundation (CREST grant #HRD-1242067), the US Department of Defense/Department of Army (Award #W911NF-11-2-0046), the US National Institutes of Health (R01 grant #CA154444), and the Komen Foundation (grant #IIR13262248).

Acquisition Time 72.318 s

Body Part Examined Chest

Slice Thickness 0.6 mm

Tube Potential 120 kVp

Tube Current 400 mAs

Exposure Time 1000 ms

Focal Spot Size 1.2 mm

Reconstruction Diameter 500 mm

Gantry/Tilt 0

ROI Size (0.72x0.72) mm

Distance from Source to Detector 1040 mm

Distance Source to Patient 570 mm

Number of Reconstructed Slice 619 (each of 512x512 pixels)