Page 61 - Fister jr., Iztok, Andrej Brodnik, Matjaž Krnc and Iztok Fister (eds.). StuCoSReC. Proceedings of the 2019 6th Student Computer Science Research Conference. Koper: University of Primorska Press, 2019

nsfer Learning Tuning Utilizing Grey Wolf Optimizer for
Identification of Brain Hemorrhage from Head CT Images
Grega Vrbancˇicˇ Milan Zorman
University of Maribor, Faculty University of Maribor, Faculty
of Electrical Engineering and of Electrical Engineering and
Computer Science Computer Science
Koroška cesta 46 Koroška cesta 46
SI-2000 Maribor, Slovenia SI-2000 Maribor, Slovenia
grega.vrbancic@um.si milan.zorman@um.si
ABSTRACT
Most commonly, diagnosing the brain hemorrhage - a con-
dition caused by a brain artery busting and causing bleed-
ing is done by medical experts identifying such pathologies
from the computer tomography (CT) images. With great
advancements in the domain of deep learning, utilizing deep
convolutional neural networks (CNN) for such tasks has al-
ready proven to achieve encouraging results. One of the ma-
jor problems of using such an approach is the need for big
labeled datasets to train such deep architectures. One of the
efficient techniques for training CNNs with smaller datasets
is transfer learning. For the efficient use of transfer learning, hemorrhages utilizing different kinds of more or less com-
many parameters are needed to be set, which are having a
great impact on the classification performance of the CNN.
Most of those parameters are commonly set based on our
previous experience or by trial and error. The proposed
method addresses the problem of tuning the transfer learn-
ing technique utilizing the nature-inspired, population-based
metaheuristic Grey Wolf Optimizer (GWO). The proposed
method was tested on a small head CT medical imaging
dataset. The results obtained from the conducted experi-
ments show that the proposed method outperforms the con-
ventional approach of parameter settings for transfer learn-
ing.
Keywords
Convolutional Neural Network, Transfer Learning, Optimiza-
tion, Biomedical images, Classification
1. INTRODUCTION
Most commonly used medical imaging technique to assess
the severity of brain hemorrhage, also termed as a cere-
bral hemorrhage, intracranial hemorrhage or intracerebral
hemorrhage is the computer tomography or shortly CT. As
reported in [24], each year intracerebral hemorrhage (ICH)
affects 2.5 per 10,000 people worldwide and is associated
with high mortality that only 38% of ICH patients could
survive over one year. Besides, more than 80% of people
are suffering due to being born with a weak spot in their
major brain arteries. However, the early diagnosis of the
condition and receiving immediate and relevant treatment
can be a lifesaver for the affected patient. Traditionally, the
tools helping in diagnosing such conditions are CT images
obtained from the CT scan, which are then examined by the
expert such as an experienced doctor, who has the ability to
identify important symptoms of the disease from the image
StuCoSReC Proceedings of the 2019 6th Student Computer Science Research Conference DOI: https://doi.org/10.26493/978-961-7055-82-5.61-66 61
Koper, Slovenia, 10 October
Vili Podgorelec
University of Maribor, Faculty
of Electrical Engineering and
Computer Science
Koroška cesta 46
SI-2000 Maribor, Slovenia
vili.podgorelec@um.si
by a naked eye [3].
With the expansion of deep learning field and with the great
achievements of deep convolutional neural networks (CNN)
for the image and video recognition tasks [26, 27] are such
approaches and methodologies also being used for addressing
various medical areas such as medical image analysis [1] and
classification [31, 12], biomedical signal segmentation [23]
and detection of various human organ activities [30].
In recent studies [12, 4, 11], the authors have already ad-
dressed the problem of identifying various kinds of brain
plex deep CNNs. However, the problem with the training
of such deep CNN architectures remains the same. In or-
der to achieve acceptable performance, the training of such
networks requires a lot of resources in terms of time and pro-
cessing power. Additionally, a big dataset of images, hand-
labeled by experts is also required. Given the fact that such
high-quality big datasets of biomedical images are hard to
obtain, researchers are trying various approaches and tech-
niques to overcome this problem. One of the most popular
techniques for training deep CNNs on small datasets is trans-
fer learning, which has already proven to achieve great re-
sults [4, 14]. But the transfer learning techniques also comes
with the downsides. Most commonly, the biggest problems
when utilizing the transfer learning approaches are finding
out which and how many layers to fine-tune and how to set
the training parameters for the fine-tuning of the CNN in
order to obtain the acceptable outcome.
Based on the encouraging results of transfer learning tech-
nique being used to train CNNs for the task of classification
of biomedical images and our previous experience on opti-
mizing various training parameters [32], we set our goal to
develop a method for an automatic optimization of trans-
fer learning utilizing nature-inspired population-based Grey
Wolf Optimizer (GWO) algorithm named GWOTLT.
The rest of the paper is organized as follows. In Section
2, we briefly describe methods which were used. In Sec-
tion 3, we present the proposed GWOTLT method, while in
Section 4 we describe the experimental setup of conducted
experiments, the results of which are presented in Section 5.
Conclusions and final remarks are gathered in Section 6.