In current generation hypoxia modified
theories have shown to provide better outcomes to cancer patients, when
compared with standard cancer treatments. The cancer treatments depends on the
proportion of the hypoxia regions (i.e a region deprived of adequate oxygen
supply) in tumour tissue so it is important to estimate this proportion. This
paper surveys on various methods that are been proposed for the detection and
quantification of tumour hypoxia using various classifier. This article also
reviews about the method where multi-modal microscopy images (ie) immuno
fluorescence (IF) and Hematoxylin eosin (HE) stained images of a histological
specimen of a tumour.
Hypoxia, Optoacoustic, Ultrasound, Cytological microscopic images,
Computational modeling, Multimodal images, Micro circulatory supply unit(MCSU).
Cancer can be defined
as a set of diseases, where the normal cell lose their controlled mechanisms in
the body and behave abnormally in the cell society. Hypoxic regions (i.e., a
region deprived of adequate oxygen supply) are commonly present in human
tumors, and they are usually associated with poor clinical prognosis 1.
Hypoxia is recognized as a factor that helps tumour cells survive by giving
them a more aggressive phenotype. Specifi- cally, it has been observed that the
efficacy of common treatments (such as standard radiotherapy, some O2-dependent
chemotherapy, photodynamic therapy, and immunotherapy) is limited in such
hypoxic regions. Hypoxia can be generally ?G.C. thanks the
Alexander von Humboldt Foundation (Fellowship for Experienced Researchers).
This work was partially supported by the Australian Research Council Centre of
Excellence for Robotic Vision (project number CE140100016). Fig.1. Manual
classification of microvessel regions from HE (top) and IF (bottom) images of
the same histological specimen and the rough mask delineating the vital tumor
region in both images. The slice represents one whole tumor cryosection, where
the pink color channel in HE denotes necrotic region and the three color
channels in IF represent three fluorescence stains (red denotes microvessels,
green displays hypoxia, and blue shows perfusion). classified into chronic or
acute 1, depending on its causes, duration and consequences 1, where
chronic hypoxia is characterized by limitations in oxygen diffusion from tumor
microvessels into surrounding tissue, while acute hypoxia is represented by
local disturbances in perfusion 1. The main result of chronic hypoxia is a
limitation of tumor growth while acute hypoxia can promote tumor aggressiveness
2. There is also evidence that fluctuating hypoxia levels with time indicates
the development of aggressive survival strategies, such as local invasion,
metastasis, and acquired treatment resistance. Therefore, a successful clinical
treatment critically depends on the use of medical imaging data for first
detecting vital and necrotic tumor tissues, and then for classifying vital
tissue regions into normoxia or hypoxia and then to further classify the
hypoxia into chronic or acute 3.
Cellular automata (CA)
models characterize the tumour cells as distinct entities of specific location
and scale, analyse the interaction among the cells and discuss the peripheral
factors in discretised time intervals with some rules which are predefined.
Multispectral optoacoustic imaging, also known as spectroscopic photoacoustic
imaging, has been widely used as an imaging technique for inferring tumour
hypoxia by visualising the distribution of oxy-haemoglobin (HbO2) and
deoxy-haemoglobin (Hb) 3, 4.Multimoda images are hematoxylin, eosin and
immune fluroscence stained images. These two images modalities are to allow the
delineation ofvital tumour tissue in both images. The immune fluroscence image
is blue in colour to view to nuclei and hematoxylin image is pink to view the
cytoplasm and extracellular particles.
The first and foremost
step in the cancer treatment is detecting vital and necrotic tumour tissues and
then classifying these tissue regions into normoxia or acute or chronic or
necrosis to make this clinical process (i e) identification and quantification
of tumour easier several methods have been proposed. this article deals with
some of the proposed methods.
and Optoacoustic tomography
This method is based on
co-registering optoacoustic tomography images with DCE-US images to demonstrate
in preclinical cancer models, the value of combining two imaging modalities.
Multispectral optocoustic imaging is an imaging technique where the hypoxia is visualized
by the distribution of oxy-haemoglobin(Hbo2) and
deoxyhaenoglobin(Hb).The total haemoglobin difficult microbubble prefusion.
The total haemoglobin (HbT) measurements estimated by optoacoustic imaging
depend on two parameters; the concentration of the optical absorbers present at
a region of interest and the system’s sensitivity to detect those chromophores
which is influenced by the spectrally dependent attenuation of light by the
tissue and (potentially) the presence of confounding chromophores that would
make the spectral recognition of Hb and/or HbO2 more difficult. Microbubble
based dynamic contrast enhanced ultrasound (DCE-US), on the other hand,
provides good information on tissue perfusion.
For optoacoustic imaging, an MSOT
inVision 256-TF was used. This consists of a optical parametric oscillator
pumped by a pulsed Nd:YAG laser, tunable for wavelengths from 710 nm to 950 nm
in steps of 10 nm (pulse duration 9 ns, repetition frequency 10 Hz). After
optoacoustic imaging, for DCE-US imaging, the anaesthetised animal in its
holder was moved to a purpose built gantry, tailored to reproduce the MSOT
imaging setup. Optoacoustic images were reconstructed using an interpolated
model matrix inversion algorithm 12. In order to assess whether the tumour
regions showing a lack of haemoglobin signal on the optoacoustic image were
perfused, regions of interest (ROIs) were drawn to compute the HbT values from
the optoacoustic image and the TICs from the DCE-US image sequences. TICs were
computed as the mean contrast signal within each ROI verses time, after
background echo image subtraction, using a program coded in Matlab (2010b,
MathWorks, Natick, MA).
For registration of the MSOT and
ultrasound images rigid body transformation was found satisfactory as there
were minimal changes to the posture and position of the anaesthetised animal
between image acquisitions. In comparison to the blood-signal regions, the
nosignal regions on the optoacoustic images had on average: a longer mean time
of arrival, time to peak and wash-in time of the microbubbles, and a lower mean
AUC, peak contrast, wash-out rate and wash-in rate of microbubbles.
The disadvantage of this method
is the registration of the MSOT and US images depends on a visual matching of
features in the two sets of images , which will vary based on the observer.
B. Multimodal cytological images
In this method mashed and
registered immune fluorescence and hematoxylin images are used to detect and
classify microvessel regions by using a combination of four classifiers. the
classifier are 1.Adaboost, 2.Linear support vector machine(linear sum), 3.Random forest and 4.Deep convolution neural networks. The
combined results is used in two ways
1. using a joint classification
from the results of four classifier.
2. using a conditional random
field(CRF) model with four unary potentials and
a binary potential that encodes contrast dependent labeling
The IF images for the tumor
cryosection were prepared with three separate stainings. Pimonidazole was used
for hypoxia stain (green regions), CD31 was used for vessel stain (red
regions), and Hoechst 33342 was used for perfusion stain (blue). Then, the
cover slip was removed to stain the same slice with HE in order to detect the
necrotic regions. This procedure can cause severe tearing and folding in HE
After staining, the whole tumor
cryosections were scanned at the same pixel size and photographed with the same
settings as the IF images. Finally, a manually delineated mask was also used in
order to remove major necrotic regions, skin, background and tissue folding and
tearing (see Fig 1). The manual labeling of the microvessel regions is
performed using an active learning scheme.
From the four classifiers the
best result is obtained from the Adaboost classifier. The disadvantage of this
method is that the results produced by CRF model does not improve over the
random forest and it is not competitive enough.
C. Cellular Automaton Model
Here a Computational model is been proposed
where the hypoxis is considered as a micro environment constraint of tumour
growth. This model uses a two dimensional cellular automata grid and artificial
neural network for establishing signaling network of tumour cells. The model
measured tumour invasion and the number of apoptotic cells to support that
hypoxia has a critical impacts on avascular tumour growth. Here the simulation
is coded using MATLAB 2013a. This provides a simulation model for tumour growt
with the effects of hypoxia. The
microenvironments include oxygen concentration, glucose concentration, cell
movement, ECM and cell-cell adhesion. It has been shown that different
microenvironments parameters strongly influence the tumour dynamics and growth.
CA method has also been used successfully for different aspects of tumour
growth modeling For this model, a 2D lattice grid has been taken into N×N.
The necrosis has been activated
when oxygen concentration goes below certain threshold level Cap. Tumour growth
started from four cells in the simulation and grew spherically in layered
structure consisting of dead region in the centre and proliferating and
quiescent cells surrounding the necrotic region. For showing the tumour hypoxia
we simulated our tumour growth model for different oxygen concentration. If we
can limit the oxygen supply during the tumour growth evolution, we can
understand the hypoxia impact clearly.
The layered structure reveals the harsh
hypoxic condition as the amount of dead cells increased and tumour shows
fingering morphology. When the lowest oxygen concentration was given during the
simulation, tumour showed the highest invasion and fingering morphology than
the growth with other concentrations. It clearly indicates that with lowest
oxygen concentration, the tumour mass create strong hypoxia conditions. The
results confirm that the lowest concentration of oxygen create a harsh hypoxic
environment and drives the tumour cells to invade the surrounding tissue
The disadvantage is that this
model did not integrate microenvironment constraint like cell – cell adhesion and cell – ECM
The objective of this paper is to show a
survey about various methods proposed for the detection and quantification of
tumour hypoxia. The implications of the cellular automaton model will be broad
and effective to the research community. These methods for quantifying the hypoxic
regions mostly provides accurate results the to the clinicians and make their
treatment process easier. In future these methods can be extended by developing
a methodology that no longer needs in-expensive annotation process.