IEEE Consumer Electronics Magazine - January/February 2020 - 62

Smart Home Healthcare

& LUNG

CANCER SCREENING is conducted during
health examinations and public lung cancer
examinations. When there is an abnormality
in these examinations, a precise examination
is performed. Computed tomography (CT) is
mainly used for the precise inspection. Since the
image captured by the CT apparatus is 3D tomographic image, detailed information about the
body can be obtained and abnormal shadows
that are hidden by a general X-ray examination
can be found. From this, it is expected that the
accuracy of diagnosis will be improved. However, as the performance of the CT apparatus
improves, the number of images obtained in one
scan becomes enormous and the number of
scans increases every year, so the burden on the
doctor who interprets the images is increasing.
This raises the concern that a lesion may be
overlooked due to a doctor's fatigue or a radiologist's inexperience.1
To solve these problems, computer-aided
diagnosis (CADx) systems are needed.2 The
CADx system uses a computer to capture images
and perform analyzes, such as segmentation,
alignment, and detection of a lesion. It results in
a "second opinion" by a doctor for diagnosing.
The improvement of image interpretation speed
and image analysis accuracy is expected when
using the CADx system because it can reduce
the burden on the interpreting physicians and
reduce variation in diagnostic accuracy.
One technology contained within the CADx
system is a temporal subtraction technique. The
temporal subtraction techniques remove normal
structures, such as blood vessels, bones, muscles,
and others, by performing difference calculation
processing between the present patient and a previous image of the same patient, highlighting only
a lesion portion that changes with time. This technique creates a differential image.3 Its efficacy has
been confirmed over time in chest X-ray images
through the clinical evaluation experiments, and,
in particular, the improvement of the correct
examination rate is recognized in the resident
physician compared to the radiologist. Therefore,
this technique is more effective for lessexperienced radiologists.4 However, there exist
line cancer, squamous cell carcinoma, large cell
carcinoma, small cell carcinoma in lung cancer,
and a frosted glass-like shadow called GGO.

This is said to be a precancerous lesion of a line
cancer in the future. As the magnitude increases
or the concentration of GGOs becomes dense,
treatment increases as the risk of malignancy
increases. Early detection of GGOs is important
for follow-up observation, but they show a faint
shadow compared to other cancers, so they are
difficult to detect early5 (see Figure 1). GGO (red
box) exists in Figures 1(a) and 1(b), but the temporal change cannot be emphasized correctly
because of its lightness in Figure1(c).
In this article, we introduce a method to reduce
artifacts in temporal subtraction images by
improving the method.5 We use supervoxel and
graph cuts6 to segment the GGOs in initial candidate regions after using the Iris filter.7 For each
extracted region, machine learning is performed to
reduce the GGO candidate regions. Finally, applying the voxel matching greatly reduces the artifacts in other nonGGO candidate regions.

STATE OF THE ART
In recent years, studies on CADx systems have
been increasing along with the development
of CT apparatuses and the high-performance
computers. In particular, the temporal subtraction technology has attracted attention because
it effectively improves the aspects of efficiency
such as accuracy of interpretation and speed.
For the temporal subtraction image generation
from a chest CT image, the current image, and a
previous image with different shooting periods
are required. However, these images will be misaligned due to the differences in the imaging
device, positioning at the time of imaging, respiration, heartbeat, etc. It is difficult to detect accurate temporal changes between the images with a
positional deviation. To solve this problem, the
image registration technology plays an important
role. The alignment technique is an elemental
technique necessary for temporal subtraction
image generation.

Image Registration Methods
Image registration refers to an operation
that compares multiple images and obtains a
conversion that associates the position in one
image with a position in the other images. The
image registration method is roughly divided

IEEE Consumer Electronics Magazine

IEEE Consumer Electronics Magazine - January/February 2020

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2020