Multimedia Tools and Applications (2022) 81:44059–44086

https://doi.org/10.1007/s11042-022-13260-w

An improved deep convolutional neural network

by using hybrid optimization algorithms to detect
and classify brain tumor using augmented MRI images

Shko M. Qader 1,2 & Bryar A. Hassan 3 & Tarik A. Rashid

4

Received: 7 August 2021 / Revised: 11 February 2022 / Accepted: 15 May 2022 /

Published online: 31 May 2022
# The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2022

Abstract
Automated brain tumor detection is becoming a highly considerable medical diagnosis
research. In recent medical diagnoses, detection and classification are highly considered to
employ machine learning and deep learning techniques. Nevertheless, the accuracy and
performance of current models need to be improved for suitable treatments. In this paper, an
improvement in deep convolutional learning is ensured by adopting enhanced optimization
algorithms, Thus, Deep Convolutional Neural Network (DCNN) based on improved Harris
Hawks Optimization (HHO), called G-HHO has been considered. This hybridization features
Grey Wolf Optimization (GWO) and HHO to give better results, limiting the convergence rate
and enhancing performance. Moreover, Otsu thresholding is adopted to segment the tumor
portion that emphasizes brain tumor detection. Experimental studies are conducted to validate
the performance of the suggested method on a total number of 2073 augmented MRI images.
The technique’s performance was ensured by comparing it with the nine existing algorithms
on huge augmented MRI images in terms of accuracy, precision, recall, f-measure, execution
time, and memory usage. The performance comparison shows that the DCNN-G-HHO is
much more successful than existing methods, especially on a scoring accuracy of 97%.
Additionally, the statistical performance analysis indicates that the suggested approach is faster
and utilizes less memory at identifying and categorizing brain tumor cancers on the MR
images. The implementation of this validation is conducted on the Python platform. The
relevant codes for the proposed approach are available at: https://github.com/bryarahassan/
DCNN-G-HHO.

* Bryar A. Hassan
[email protected]

1
Information Technology Department, University College of Goizha, Sulaimani 46001, Iraq
2
Department of Information Technology, Computer Science Institute, Sulaimani Polytechnic
University, Sulaimani 46001, Iraq
3
Department of Information Technology, Kurdistan Institution for Strategic Studies and Scientific
Research, Sulaimani, Iraq
4
Computer Science and Engineering Department, University of Kurdistan Hewler, Erbil, Iraq
44060 Multimedia Tools and Applications (2022) 81:44059–44086

Keywords Brain tumor . Medical MRI imaging . Deep convolutional neural network . Harris
hawks optimization . Grey wolf optimization . Medical diagnosis

1 Introduction

Brain tumors are considered a dangerous and scrupulous disease. Tumors can be caused by the
abandoned growth of the cancerous cells because they have a critical structural part containing
50–100 trillion neurons. Therefore, diagnosing brain diseases becomes complicated because of
the presence of the skull around the brain. The stages of the risk involved in a brain tumor
depend on the following factors: the size of the tumor, style of the tumor, behavior, position,
and status of growth level [41]. Brain tumors are principally classified into two types that are
malignant and benign [61]. The tumor, which does not have cancerous cells, is less injurious to
humans is known as benign. The tumor contains cancerous cells, which are more harmful to
humans than malignant [30].
The process of diagnosing the brain tumor becomes tedious. Some imaging processes
regularly do the diagnosing process, these are; single-photon Emission Computerized Tomog-
raphy (SPECT), Positron Emission Tomography (PET), Computer Tomography scan (CT
scan), Magnetic Resonance Imaging (MRI), magnetic resonance spectroscopy (MRS), and
Functional Magnetic Resonance Imaging (Functional MRI scan) test similar to the spinal tab,
biopsy and Angiogram [52]. To generate images, CT scans utilize the X-ray. CT scan exposes
high ionizing radiation to the patients while scanning [39]. This process will lead to an increase
in the risk of brain tumors. MRI provides in-depth clear images and often prescribed tests to
detect brain tumors because it is a noninvasive approach [12]. MRI uses radio waves and
magnetic fields to provide brain images. MR images offer more precise and complete images
than CT scans [4]. Therefore, MR imaging is the preferable screening investigation for
diagnosing brain tumors since the segmentation process of brain tumors is useful for identi-
fying the growth rate and organizing the clinical procedures. MR images of the brain involves
more robust procedures, and it is not to be harmful that are assessment of tissue processing,
imaging, and analysis, metabolism, physiology, and function. MRI outputs are expanding the
knowledge in medical research to improve the study of the structure of organisms’ normal and
diseased parts [50]. Brain segmentation includes segregating the different tumor cells into the
effective tumor, solid, edema, and necrosis from the normal brain cells, these are GM, WM,
and CSF. GM denotes Grey Matter, WM denotes White Matter, and CSF represents the
cerebrospinal fluid [6]. Typically, MRI has noises caused by the operator’s performance,
equipment, and circumference, which would lead to an inaccurate result. The noises presented
in the MR images can be eliminated by the denoising approaches, such as Stationary Wavelet
Transform (SWT), sharpening filter, and median filter. To detect the brain tumor, exact
segmentation becomes a critical process. The manual segmentation process requires more
time to detect the brain tumor, as a result, automated or semi-automated approaches are needed
to detect the tumor accurately. Different kinds of machine learning approaches have been
utilized for identifying the tumor through the MRI along with the k-means, thresholding,
Fuzzy C-means (FCM), level set approaches, and kernel extreme learning machine (KELM)
[3, 33]. Even though machine learning techniques have some issues, such as interpretation of
results, high error susceptibility [40]. To overcome the issues, deep learning-based approaches
are creating more attention. Deep learning methods are preferred for the segmentation process,
particularly CNN is best for recognizing patterns [38]. Moreover, these approaches absorb the
Multimedia Tools and Applications (2022) 81:44059–44086 44061

features in terms of hierarchy while comparing with the other statistical methods, including
support vector machine (SVM); these depend on the hand-crafted features [31]. Deep CNN
approaches are successfully implemented for analyzing the clinical images with the retrieval,
segmentation, and classification [48].
Based on the previous works presented above, the following are the difficulties of brain
tumor detection encountered by current methodologies:

& The first disadvantage of the current models is their binary categorization of tumors,
creating further ambiguity for the radiologist. Additionally, a lack of data hinders the
researcher’s inability to get reliable results [51].
& Tumor categorization using brain imaging is a challenging task for two reasons. The first
reason is that brain tumors exhibit a great degree of variability in size, severity, and form.
The second reason is that tumors come in various pathological varieties, showing the same
symptoms [9].
& In [43], a method called Adaptive Convex Region Contour (ACRC) was developed for
segmenting brain tumors using magnetic resonance imaging. The SVM was used to
categorize the slice to determine whether it was expected or pathological. Furthermore,
the technique demonstrated enhanced performance and resolved high dimension, short
sample size, and nonlinearity problems. However, this method required tremendous
feature values, introducing noise into the classification results.
& In [54], a convolutional neural network-based deep learning model was developed to
identify different brain tumors. While deep learning methods improve tumor classification,
they need a large amount of training data for analysis. Moreover, the computational cost
and training time associated with brain tumor classification are substantial.
& Segmentation and identification of tumorous areas from brain MRIs are complicated and
take much time for analysis. In addition, the method’s accuracy is significantly impacted
when identifying brain tumors [45].

The images used to identify brain tumors are important since brain tumors are considered the
most severe illness. Accurate diagnosis of a brain tumor may enable the tumor’s precise
location in the brain to be determined. As a result, the mortality rate will decrease. Thus, it is
critical to detect hidden patterns to improve images and the diagnostic process. However,
obtaining an appropriate diagnosis for the different lesion types becomes difficult, regarded as
the primary issue. To address this issue, traditional methods are employed, including present-
ing a DCNN based on the enhanced HHO methodology. This study aims to provide a method
for detecting brain tumors that incorporate a DCNN presentation based on the improved HHO.
The proposed approach automatically detects brain cancers in magnetic resonance images,
with feature extraction doing the classification. To begin, magnetic resonance images of the
brain are regarded as input for pre-processing to create images suitable for subsequent
processing. The pre-processed images are then subjected to the segmentation process using
the Otsu thresholding method. Once the segmentation procedure is complete, extracting the
texture and statistical characteristics from each segment could be carried out. The character-
istics include the tumor’s size, variance, and mean. Moreover, these retrieved features are
formulated in the feature vector. Finally, the detection of brain tumors is carried out using
DCNN and the feature vector, where the Deep CNN method is trained using improved HHO.
The suggested approach enhanced HHO combines the standard GWO and HHO optimizations
to benefit practical classifier training.
44062 Multimedia Tools and Applications (2022) 81:44059–44086

This research is distinguished from earlier ones by the following significant contributions:

& G-HHO improves DCNN to detect and classify brain tumors from augmented MRI images
effectively;
& The G-HHO has the features of GWO and HHO;
& Otsu thresholding is used to segment the tumor part, emphasizing the identification of
brain tumors;
& Experiments are conducted by the Python platform on vast augmented MRI image
datasets;
& The performance of the suggested DCNN-G-HHO is improved and enabled better detec-
tion and classification results.

The remainder of this research paper is ordered as follows: Section 2 describes the traditional
brain tumor detection methods used in the literature and the difficulties encountered, which
served as the motivation for creating the suggested approach. Section 3 describes the recom-
mended technique for detecting brain tumors using DCNN based on improved HHO.
Section 4 proposes G-HHO to improve DCNN to classify brain tumors. Experimental results
are also discussed in Section 5. Section 6 compares the suggested approach to alternative
ways, and Section 7 concludes remarks of this work.

2 Literature review

Brain tumors are a particularly aggressive kind of cancer that may result in significant
problems in the human body. Thus, early, and accurate identification of a brain tumor may
significantly improve the chance of survival. However, proper identification of various tumor
types is a difficult problem. Thus, developing an efficient tumor representation via an
optimization algorithm is critical toward successful brain tumor identification. Ten current
methods for detecting brain tumors are examined in this research work, and the shortcomings
of each approach serve as inspiration for developing a new brain tumor detection strategy.
Thaha MM et al. [58] have presented the enhanced convolutional neural network for the
computerized segmentation process using the BAT algorithm. Furtherly, they have used skull
stripping and image enhancement algorithms for the pre-processing. Chen S et al. [8] have
presented the segmentation models, such as multi-level deep medic, dual force training
scheme, auxiliary classifier, and multi-layer perceptron-based post-processing approach to
improving the standard of the hierarchical features for detecting brain tumors. Nonetheless,
they failed to provide enhanced segmentation capability for the deep architecture to avoid the
individual post-processing steps. Anaraki AK., Ayati M., and Kazemi F., [5] have presented
the hybrid method based on the Genetic Algorithm (GA) and convolutional neural network
(CNN) to categorize the various category of glioma by MRI. Furtherly, an ensemble algorithm
was used to reduce the prediction error variance. Talo M et al. [55] have presented pre-trained
models for automatically classifying MR images, such as AlexNet, egg-16, ResNet −18,
ResNet- 34, and ResNet-50. The images are categorized into normal, cerebrovascular, neo-
plastic, degenerative, and inflammatory diseases classes. Toğaçar M et al. [59] have presented
the convolutional neural network model called BrainMRNet based on the attention modules
using the augmentation and hyper column approaches. Where Accessible magnetic resonance
images were utilized to identify the brain tumor along with the brainMRNet model. Özyurt F
Multimedia Tools and Applications (2022) 81:44059–44086 44063

et al. [42] have presented the hybrid method of Super-resolution fuzzy-c-means CNN (SR-
FCM-CNN) based on the convolutional neural network, and the learning machine algorithm,
and fuzzy c-means to identify the brain tumor. Where features are extracted by the squezeeNet
architecture from the CNN model. Sharif M et al. [53] have presented the brain surface
extraction method to get skull removed images. They have presented particle swarm optimi-
zation (PSO) for segmentation purposes. They have used GA for feature selection. Also,
artificial neural networks and classifiers were used to categorize the tumor levels. Navid
Ghassemi et al. [11] have presented the deep learning technique to classify brain tumors in
MRI images. Where deep neural network was pre-trained in a generative adversarial network
(GAN) to separate the features and analyze the structure of MR image within its convolutional
layer. Zeineldin RA et al. [62] have presented the Deepseg technique from the genetic deep
learning architecture to provide computerized identification and segmentation of tumors
through the FLAIR MRI data. For the extraction process, they were used CNN. The CNN
models, such as dense convolutional network (DenseNet), NASNet, and residual neural
network (ResNet) were used based on the improved architecture of U-Net. Vijh S et al. [60]
have presented the particle swarm optimization with the Otsu algorithm for automatically
diagnosing the brain tumor. An anisotropic diffusion filter was used to eliminate the noise, and
a convolutional neural network was used for the classification process. Finally, the most recent
research work has presented an optimization-based approach, called Whale Harris Hawks
optimization (WHHO), for brain tumor diagnosis utilizing magnetic resonance imaging (MR)
images [45]. Segmentation was carried out utilizing cellular automata and rough set theory,
and several parameters were retrieved from the segments. Furthermore, brain tumor diagnosis
was accomplished via DCNN, which was trained using the proposed WHHO. The suggested
WHHO method was created by combining the Whale Optimization Algorithm (WOA) with
the HHO. With a maximum accuracy of 0.816, a maximum specificity of 0.791, and a
maximum sensitivity of 0.974, the suggested WHHO-based DeepCNN outperformed previous
approaches (Table 1).
Based on the above-presented table, current methods’ difficulties are (i) Insufficient
precision in identifying brain tumors. (ii) Inaccurate classification of brain tumors. (iii) Lack
of using large MRI image datasets as input data. Therefore, this paper provides the following
contributions for mitigating the limitations of the current algorithms:

& G-HHO enhances DCNN to identify and classify brain malignancies from enhanced
magnetic resonance imaging pictures more efficiently.
& G-HHO provides all the good characteristics of both GWO and HHO.
& Otsu thresholding is used to segment the tumor portion, emphasizing the detection of brain
tumors.

Additionally, Table 2 indicates the primary questions of this work and how we have addressed
them.

3 The proposed methodology

During brain tumor detection, the accuracy level of the images is considered an essential one
because brain tumors are considered the most critical disease. Accurate diagnosing of a brain
tumor may identify the exact affected portion of the brain. Thus, it will reduce the death rate.
 44064

Table 1 Summary of current methods for brain tumor detection.

Reference, author(s) The algorithm Concluding remarks Limitations

[58], Chen S et al. Dual-force convolutional The proposed approach promoted the segmentation Failed to provide enhanced segmentation capability for the deep
neural networks performance of BRATS 2017 and BRATS 2015 architecture.
datasets.
[8], Thaha MM et al. CNN CNN results are promising in BRATS 2013 and The CNN components should be further investigated to develop
2015 datasets in terms of accuracy and a more robust algorithm for brain tumor segmentation using
computational time. MRI data.
[55], Talo M et al. CNN CNN obtained the best classification accuracy The fewer available annotated data.
among the five pre-trained models.
[5], Anaraki AK, and CNN based on GA The findings demonstrated the suggested method’s The proposed algorithm should be applied on larger datasets with
Ayati M, Kazemi F efficacy in classifying brain tumors using MRI several tumor types and other CNN structures and deep learning
images. algorithms to improve performance.
[59], Fatih Özyurt et al. BrainMRNet BrainMRNet was more successful than the current BrainMRNet should be used on different medical images and in
CNN models (AlexNet, VGG-16, and various fields.
GoogleNet).
[42], Özyurt F et al. SR-FCM-CNN) SR-FCM-CNN determined that brain tumors have The performance of SR-FCM-CNN varies depending on the training
been better segmented and removed using the dataset.
SR-FCM method.
[53], Sharif M et al. Ensemble algorithm The suggested approach performed better in The ensembled algorithm can be extended for the detection
comparison with existing methods. of sub-structures of a tumor.
[62], Zeineldin RA et al. DeepSeg DeepSeg showed successful feasibility and The validation of the DeepSeg framework should be extended
comparative performance for automated brain on more image datasets from other different MRI modalities.
tumor segmentation in FLAIR MR images.
[60], Vijh S Hybrid OTSU+APSO The suggested algorithm yielded a 98% accuracy The model should be further optimized for various data modalities,
rate, superior to current methodologies. and additional metaheuristic algorithms will be utilized to enhance
the diagnosed system’s performance.
[45], Rammurthy D, and WHHO-based DeepCNN The WHHO-based DeepCNN outperformed other Additional brain tumor datasets should be utilized to maximize the
Mahesh PK algorithms with maximal accuracy, specificity, suggested method’s efficiency. Also, several optimization
and sensitivity. approaches will be investigated to improve the efficiency of the
introduced method.
Multimedia Tools and Applications (2022) 81:44059–44086
Multimedia Tools and Applications (2022) 81:44059–44086 44065

Table 2 The research questions and mitigations of this work.

# Research question Mitigation

1 Inadequate accuracy in the identification of brain tumor Increasing sufficient detection of brain tumors.
2 Inaccurate brain tumor categorization Increasing accuracy of classifying brain tumors.
3 A deficiency in the use of big MRI image datasets as input Augmented MRI images are used as the input
data dataset.

Therefore, it is essential to identify the hidden patterns to enhance the quality of images and
enhance the diagnosis process. But attaining accurate diagnosis for the various lesion cases
becomes a complex one, considered the main problem. To overcome this problem, conven-
tional techniques are used, such as a DCNN based on the improved HHO technique, are
presented. The presented model automatically identifies brain tumors from the MR images,
where feature extraction executes the classification. Firstly, MR images of the brain are
considered input for pre-processing to make the images for further processing. After that,
the pre-processed images are induced to the process of segmentation through the Otsu
thresholding technique. Once the segmentation process is completed, then the features in
every segment as per texture and statistical are extracted. The features include the size of the
tumor, variance, and mean. Furtherly, in the feature vector, these extracted features are
formulated. At last, by DCNN with the feature vector, brain tumor identification is performed
where the presented enhanced HHO trains the Deep CNN technique. The proposed technique
of improved HHO combines the conventional GWO and HHO to get the advantages of the
two optimizations for effectual training the classifier. The overall execution process of the
proposed DCNN-G-HHO is shown in Fig. 1.

Fig. 1 The proposed DCNN-G-HHO model for brain tumor detection.

44066 Multimedia Tools and Applications (2022) 81:44059–44086

Consider the D database and the g number of input MRI images, and it can be mathemat-
ically represented in Eq. (1).


D ¼ I 1 ; I 2 ; ……:I e ; ……::; I g ð1Þ

Here, Ie denotes the eth image and g denote the MRI image’s total count in the Database.
In the following sub-sections, three components (Pre-processing, Segmentation using Otsu
threshold, and feature extraction) of the proposed DCNN-G-HHO are detailed.
The proposed method consisted of the following stages:

3.1 Pre-processing

The primary usage of pre-processing is providing a further smooth process of the input image.
The pre-processing step is more required to make the images appropriate for the identification
process for processing the image. Furtherly, the pre-processing is executed to remove the
artifacts and noises presented in the extracted images. Thus, the pre-processing step performs
image enhancement to improve the contrast of the images to detect the brain tumor. Then, for
the segmentation process, the pre-processed images are given for fetching the essential features
which are more suitable for identifying the tumor.

3.2 Segmentation

Using the cellular automata model, the pre-processed images are given to the segmentation
module to produce segments. The pre-processed images contain various elements, and every
segment denotes the separate regions. In brain tumor identification, Otsu thresholding is
employed to discover the cancerous portions by each component.
Otsu algorithm can be distinguished among the foreground and background of the images
by setting the image above the calculated value of grayscale. Before executing the segmen-
tation algorithm, the Otsu algorithm needs to calculate the image’s grey level histogram. The
primary purpose of the Otsu technique is to discover the threshold that reduces the intra-class
variance, and it can be described as the two-class weighted sum of variances as shown in Eq.
(2) [1].

σ2ω ¼ ω0 ðt Þσ20 ðt Þ þ ω1 ðtÞσ21 ðtÞ ð2Þ

Where, ω0 and ω1 represent the weights that are the possibilities of the classes, and by the
threshold value t, these weights are separated. σ20 and σ20 represent the variances of two classes.
After that, the images are binarized more than the threshold value, and covering the holes
presented in the segmented image can be made by executing the morphological operations.
Furtherly the segmented images are masked upon the original images; thus, the segmented
tumor is attained.

3.3 Feature extraction

By now, numerous research on deep feature extraction approaches has been conducted, with
promising results for image classification. One of the feature extraction models was proposed
by [56] to combine both approaches into a unified framework and to develop an unsupervised
linear feature selective projection (FSP) with low-rank embedding and dual Laplacian
Multimedia Tools and Applications (2022) 81:44059–44086 44067

regularization exploit the intrinsic relationship between data and suppress the impact of noise.
The proposed approach outperformed existing state-of-the-art algorithms for feature extraction
in the presence of different types of noise. Another study conducted by [29] for feature
extraction, took into consideration both the minimizing of reconstruction error and the
maximizing of variance in a unified framework. The experimental findings from multiple
databases demonstrated the model’s efficacy. However, it was expected to explore a more
straightforward but efficient method for its selection. Last but not least, unsupervised multi-
view feature selection using cross-view local structure-maintained diversity and consensus
representation learning was suggested by [57]. Projecting each original data view onto a shared
label space with a consensus component and a diversity part captures shared and distinct
knowledge across images. Numerous pre-defined view-specific similarity graphs are employed
to train a shared similarity network across views. Despite its shortcomings, the suggested
method’s effectiveness is shown by a parameter sensitivity analysis on real-world Multiview
datasets.
To address the limitation of the above studies, the feature is obtained from the MRI input
images to create a feature vector. When attaining the segments, the features are extracted, and
it carries out the pre-processed images for every segment. The feature extraction assures the
effectual identification of brain tumors because it involves texture and statistical features. For
extracting the features, every segment considers it for attaining better accuracy in detecting
brain tumors. As earlier discussed, from each segment the feature extraction proceeds in terms
of tumor size, mean, variance. This can be described below:

3.3.1 Mean

The mean value can be computed by summarizing the entire numbers presented in the dataset
and dividing it by the total count of the number presented in the dataset. The mean value is
considered the vital parameter for the image segmentation process. A two-layer feed-forward
neural network can forecast the growth of the tumor within the 16% MSE. The mean can be
computed by calculating the pixel’s average presented in the image, and it can be mathemat-
ically expressed as formulated in Eq. (3).
1 jd ðS Þj
μ¼  ∑n¼1n d ðS n Þ ð3Þ
jd ðS n Þ

Here, n represents the total segments, d(Sn) denotes the every segments pixel values and
∣d(Sn)∣ represents the total pixel of the segment.

3.3.2 Variance

Measuring the spread among the numbers of a dataset is known as a variance. This variability
implies measuring the distance between each number from another number in a dataset.
Variance can find the reasons for the deviation in the brain, and it helps recognize the tumor
cells in the brain. The variance features are computed and mathematically expressed in Eq. (4)
according to the mean value.
jd ðS Þ
∑n¼1n jS n −μj
σ¼ ð4Þ
d ðS n Þ
44068 Multimedia Tools and Applications (2022) 81:44059–44086

3.3.3 Tumor size

From the Brain MR images, tumor size extracts the features to detect the brain tumor. Tumor
segmentation process from the brain MRI contains various covering pathology such as MRI material
science, perception of radiologist and analysis of the image. A brain tumor can be of any size; it has
different shapes and may locate on any portion and may show deviation in the image intensities.
The variable Q indicates the tumor affected boundary, the derivation of the area of feature
vector from the tumor-dominated portion. Q is mathematically represented in Eq. (5).
π
Q¼ ðL  W Þ ð5Þ
4
Where L represents the length of the tumor size, W denotes the width of the tumor size.

3.3.4 Formation of the feature vector

The below expression describes the set of texture and statistical features. Therefore, the
extracted features of every segment are represented as shown in Eq. (6).
J ¼ fμ; σ; ε; μg ð6Þ

Where J represents the extracted feature vector by each segment such as μ, σ, ε, μ denotes the
mean. σ denotes the variances. ε, indicate the tumor size. After that, the extracted feature
vector is fed to the DCNN, which categorizes the images based on the given features, and
forms the class label. The classifiers are used to formulate the class label and segregate the
cancerous and noncancerous cells concerning the input image.

4 Proposed G-HHO to improve DCNN

This section, presents the proposed G-HHO technique to detect the brain tumor where the
detection process is executed with the consideration of a feature vector. Moreover, the extracted
features are further processed for classification by DCNN trained with enhanced HHO to classify
the tumor. In the improved HHO, the parameters of GWO are utilized; therefore, the performance
of HHO can be improved. Based on the social behavior of the GWO is developed. The GWO
algorithm is more suitable to get optimum global results because it gives a better convergence rate
[10]. Moreover, GWO can resolve real-time problems and complicated search spaces. Also, HHO
is adopted from Harris’s hawks’ helping and chasing behavior. This approach is more effective
for witnessing optimization problems and providing optimum results. Even though the HHO
technique has some issues in the search space, such as multi-modality, deceptive optima, and
optimal local solution, the proposed work that combines the GWO with HHO (G-HHO) will give
better results it limits the convergences rate and enhancing the performance.
Furthermore, our proposed hybrid approach provides better exploration and exploitation
than other meta-heuristic algorithms, such as minimal sequential optimization (SMO), BAT,
and PSO. The exploration and exploitation of search areas are calibrated to emphasize
exploitation as the iteration counter grows. The suggested DCNN-G-HHO model incorporates
a trade-off between the exploitation and exploration phases to identify the optimal solutions
and converge to the global optimum. Regardless of the common features among the current
meta-heuristic algorithms, The DCNN-G-HHO adds a feature of improved searching spaces.
Multimedia Tools and Applications (2022) 81:44059–44086 44069

The search processes are divided into exploration (diversification) and exploitation (intensifi-
cation). The algorithm should maximize the use and promotion of its randomized operators
throughout the exploration phase to examine different areas and sides of the feature space
thoroughly. Thus, the exploratory behaviors of a well-designed optimizer should be sufficient-
ly random to efficiently distribute more randomly produced solutions around the problem
topography during the early stages of the search process [27]. Typically, the exploitation step
follows the exploration stage. The optimizer concentrates on the neighborhood of higher-
quality solutions situated inside the feature space during this phase. It accelerates the search
process inside a specific location rather than throughout the terrain. A well-organized opti-
mizer should strike a sensible and delicate balance between exploration and exploitation
inclinations. Otherwise, the likelihood of being caught in local optima and suffering from
immature convergence downsides grows.
The algorithmic stages of suggested improved HHO and the structure of the Deep CNN
model are defined in Eq. (7).
T ¼ fT 1 ; T 2 ; …:; T h ; …:T l g ð7Þ

Here, l denotes the total count of the convolutional layer of the Deep CNN model and hth a
convolutional layer of the Deep CNN can be represented by the Th. Using the units (p,w), the
output can be derived and expressed in Eq. (8).
 h      
W P−1 ru1 ru2
T ϑ p;w ¼ vhϑ p;w þ ∑p¼1 1
∑z¼−r u ∑
s¼−r
u
u X f ;p * T h−1
v pþz; wþs
ð8Þ
1 2 z;s

Where the conv operator can be denoted as * and is used for permitting the extraction process
of pattern from the results taken out from the adjacent convolutional layer, feature maps can be
 
represented as j T u−1
p j . total count of feature maps can be denoted by W P−1
1 and
  mþz; rþs

X uf ;p Denotes the weights. Where the trained by the proposed G-HHO.

z;s
The following three sub-sections are given below for the proposed algorithm (G-HHO).

4.1 Rectified linear unit and pooling layer

Rectified linear unit and pooling layer (ReLU) is considered an activation function because its
contribution is highly significant in simplicity and efficiency. Moreover, ReLu in Deep CNN
provides speed processing with large networks. The output obtained through the ReLu layer is
given based on providing feature maps in Eq. (9).

T uf ¼ fun T u−1
f ð9Þ

Where T uf represents the input and T u−1

f represents the output and fun () represents the uth layer
activation function.

4.2 Fully connected layers

To initialize the classification process of objects, the patterns of the convolutional and pooling
layers are provided as an input to the completely connected layers. The output of the
completely corresponding layer can be mathematically expressed in Eq. (10).
44070 Multimedia Tools and Applications (2022) 81:44059–44086
     
W P−1 W P−1 W P−1
S uf ¼ Z auf where auf ¼ ∑p¼1
1
∑m¼1
2
∑ n¼1
3
vuf ;p;m;n : T u−1
f ð10Þ
m;n

where, V uf ;p;m;n represents the weights (m,n), which is connected with the u − 1 layer’s pth
feature map and u layer’s fth unit. By the proposed G-HHO approach, these weights are
optimally tuned.

4.3 The network architecture of the proposed DCNN model

The proposed work has six architecture layers that incorporate three convolutional layers, two fully
connected layers, and the last layer, one classification layer. These convolutional layers are connected
by three rectified linear operator (Relu) layers and max-pooling layers. The initial convolutional layer
has a total number of fifty-two filters with a size value of 7 × 7 pixels, a stride of 2 pixels, and no
padding is coming after ReLu. A max-pooling layer has a value of 3 × 3 region and 2 pixels for
stride. The result obtained through convolutional layer 1 is 52 × 69 × 69; this result is considered an
input value for the next layer. The second convolutional layer has filters 256, and the size of them are
5 × 5 pixels, two pixels for stride, and similar to the first convolutional layer, no padding is set after
Rectified Linear unit as well as ReLu, and 3 × 3 regions as a value of max pooling and 2 pixels for
stride. The resulting outcome for the second convolutional layer has obtained 256 × 15 × 15 pixels
as size as the first convolutional layer. This result is taken as an input to the third convolutional layer.
The third convolutional layer is set with 156 filters with a 3 × 3 region and 2 pixels for strides. The
third layer result goes to the fourth layer as input; this layer is the fully connected layer with neurons
512. This layer comes after the dropout layer with a probability value of 0.5. At last, the classification
layer shows the results with two classes. Using appropriate techniques to tune the hyperparameters of
DCNN is significant to improve the classification technique [47]; therefore, the combination of Grey
Wolf Optimization and Harris Hawks Optimization is considered to enhance the performance. The
following section illustrates the GWO and HHO in detail.

4.3.1 HHO

HHO technique is considered to enhance the performance of the deep learning technique [28].
HHO is mainly relying on two stages exploration and exploitation. As represented in Fig. 2,
the following section provides the step-by-step procedure of HHO.

4.3.2 Exploration

The first stage is the exploration stage, where the entire Harris Haws represent the candidate
solution. Here, fitness value is based on targeted prey that may change every iteration.
Equation (11) represents the exploration performance [28].

X ðiÞ−r1 jX rsolu
 ðiÞ−2r2 X ðiÞj q≥ 0:5 rule ðiÞ
X ði þ 1Þ ¼  rsolu ð11Þ
X prey ðiÞ−X m ðiÞ −r3 ðLB þ r4 ðUB−LBÞÞ q < 0:5 rule ðiiÞ
Where:
X(t + 1) => Represents the position of Hawk’s in the next iteration i.,
i = > Iteration,
Xprey = > Prey’s position,
Xrsolu(i) = > Random solution selection process based on current population,
X(i) = > Represents the position vector of Hawk’s based on current iteration,
Multimedia Tools and Applications (2022) 81:44059–44086 44071

Fig. 2 Various stages of Harris Hawks Optimization (Extracted from [11]).

r1, r2, r3, r4, and q = > Random scaled factor within range of [0, 1] (updated in every
iteration),
LB => Lower bound of variables,
UB => Upper bound of variables,
Xm = > Solution’s average number.
Based on two rules, the position of hawks is developed by this targeting strategy within
(UB − LB).
Two strategies are adopted to determine the Hawks’ position (UB-LB). The initial rule is
Hawks based solutions are generated randomly based on the current population and other
hawks. The second rule is a solution prepared based on the Hawk’s average position, prey’s
location, and the random scaling factor. These steps help to enhance the randomness of the
rule. The randomly scaled action in terms of length is fed to the Lower Bound (LB). Therefore,
various feature spaces can be explored, and the random scaled feature enables the diversifi-
cation technique. Equation (12) represents the average solution position [28].
1 N
X m ðiÞ ¼ ∑ X i ðiÞ ð12Þ
N i¼1
Where:
Xm(i) = > Coordinate average of solutions which is presented in the current iteration,
N = > Whole possible results,
Xi(i) = > Each result’s location in ith iteration,
44072 Multimedia Tools and Applications (2022) 81:44059–44086

The Hawk’s prey catching process is based on a random solution that can be achieved
through Eq. (10 (i)). The entire hawks show the best solution can be done through Eq. (10 (ii)).

4.3.3 Transition from exploration to exploitation

Based on the rabbit’s energy (E), the movement of Harris hawk’s optimization from explora-
tion to exploitation is described in this stage.
In this algorithm, the prey (Eng) energy is significant to execute the optimization through
the exploration and exploitation shown in the following Eq. (13) [28]. The preys escaping
pattern will be reduced due to the action of HHO. The energy reduction is indicated from [1,
−1] is indicated as Eng0.
i
Eng ¼ 2Eng0 1− ; Eng0 ∈½−1; 1 ð13Þ
I
Where:
I = > Overall iterations,
i = > Current iteration.

4.3.4 Exploitation

At the time of hunting, the prey takes maximum effort to escape from being attacked.
Therefore, HHO executes four different techniques to confuse its prey in the exploitation
stage, where the position is highly significant. These techniques are soft besiege, soft besiege
with progressive rapid dives, hard besiege, and hard besiege with advanced quick dives.
Based on variables such as r and |Eng|. That is escaping probability and prey’s energy level,
respectively. These two variables are essential in the HHO strategy. The condition applied is r <
0.5, which shows the higher possibility for prey escaping. The r ≥ 0.5 condition indicates chances
of running are reduced. The following sections provide brief illustrations about each tactic.
a) Soft besiege.
Rabbit is the usual prey of Hawks which has adequate energy level to escape from being
attacked that can be illustrated by |Eng| ≥ 0.5 and r ≥ 0.5. On the other hand, Hawks executes
its tactics to minimize its prey energy level by encircling it by soft besieges. This strategy is
done before running the surprise pounce of the Hawk. The mathematical derivation is shown
in Eq. (14) below [28]. 8
< X ðt þ 1Þ ¼ ΔX ðtÞ−E jJX rabbit −X ðtÞj
ΔX ðtÞ ¼ X rabbit −X ðtÞ ð14Þ
:
J ¼ 2ð1−r5 Þ; r5 ∈½0; 1
Where:
J => Prey’s jump power,
r5 = > Random variables,
X(i + 1) => Hawk’s position in the next iteration of i,
X(i) = > Hawk’s position vector in the current iteration i,
Eng = > Energy of prey,
ΔX(i) => Variations between the prey’s position vector and current location in iteration i.
b) Hard besiege.
Here, the rabbit is got tired due to the soft besiege, and its energy is reduced, where the
values of the variables are |E| < 0.5 and ≥0.5.
Multimedia Tools and Applications (2022) 81:44059–44086 44073

The soft besiege makes the prey tired (energy level reduced), where the values of variables
are |Eng| < 0.5 and ≥0.5. Then, hawks process the surprise pounce hardly and encircle the
prey. The derivation is shown in Eq. (15) [28].
X ðt þ 1Þ ¼ X rabbit ðt Þ−E jΔX ðtÞj ð15Þ

c) Soft besiege with rapid dives.

In this stage, the prey still has some energy for escape with the variable’s value of |E| ≥ 0.5
and r < 0.5. Therefore, the Hawk is smartly encircling the rabbit, and before the performance of
surprise pounce, it patiently dives. Here, two steps are utilized to update the hawks’ position,
and the event of dive is known as intelligent soft besiege. The Hawk moves forward to the rabbit
with the help of the prey’s next move. Equation (16) [28] is derived from the action above.

Y ¼ X prey ðiÞ−Eng JX prey ðiÞ−X ðiÞ ð16Þ

Then, the possible results are compared based on movements executed at earlier dive to decide
whether it is a good dive.
The Hawk generates irregular dives when decisions are not taken based on the levy flight
(LF) method. This generation is formulated in Eq. (17) [28].
Z ¼ Y þ S  LF ðDimÞ ð17Þ

Where:
Dim = > Dimension of solutions,
S = > Random vector of size 1 × dim,
LF = > Levy Flight function.
The levy flight function is calculated by using Eq. (18) [28].
  ! β1
uσ Γð1 þ β Þ  sin πβ
LF ðxÞ ¼ 0:01  ; σ¼    β−1
2
 ð18Þ
Γ 1þβ  β 
1
jvj β 2 2 2

Where:
β = > Default constant which is automatically set as 1.5,
u, v = > Random values within [0,1].
Hence, the position of the Harris hawks is updated with the progressive rapid dives
calculated using Eq. (19) [28].

Y if F ðY Þ < F ðX ðtÞÞ
X ðt þ 1Þ ¼ ð19Þ
Z if F ðZ Þ < F ðX ðtÞÞ
Where:
Y and Z = > Next location of the new iteration t.
This new iteration is performed by utilizing the eqs. (18) and (19).
d) Hard besiege with progressive rapid dives.
Here, the values of variables are |Eng| < 0.5 and r < 0.5. In this situation, hawks try to
reach the rabbit by rapid dives because it has no more energy to escape. This action is
performed before the surprise pounce to catch the rabbit. In this Equation, Z is updated by
utilizing Eq. (20), and Y is accomplished by using Eq. (20) [28].
Y ¼ X prey ðiÞ−E JX prey ðiÞ−X m ðiÞ ð20Þ
The Xm(i) is obtained by utilizing Eq. (12), the four strategies are helping hawks catch their prey.
44074 Multimedia Tools and Applications (2022) 81:44059–44086

With the help of exploration and exploitation techniques of Harris Hawks Optimization, the
weightage of the classification technique’s parameters is enhanced in the proposed work.

4.3.5 The proposed G-HHO for brain tumor detection

By utilizing the presented enhanced HHO algorithm, the DCNN training process can find the
optimum weights for tuning the DCNN classifier. The hyperparameters tuning is done by
adopting GWO [2] and HHO. The HHO is improved by utilizing the technique GWO; therefore,
the performance of the deep learning technique can be significantly enhanced on performance
level. The proposed work has implemented the HHO strategy to select the optimization problem
in choosing the hyper-parameters. This algorithm enables a better searching strategy; therefore,
the performance and results can be most effective. As discussed in the previous section, DCNN
performance is improved by tuning the parameters of CNN with the G-HHO strategy. The
parameters utilized for this work are three convolutional Layers, two fully connected layers, and
finally, one classification layer. Each convolutional layer is connected by three ReLu layers and a
max-pooling layer. The pseudocode of the proposed method is presented in Algorithm (1):
Algorithm 1: Pseudocode of the Proposed G-HHO.
Multimedia Tools and Applications (2022) 81:44059–44086 44075
44076 Multimedia Tools and Applications (2022) 81:44059–44086

5 Experimental setup

This section integrated with the proposed brain tumor detection outcome through an experimental
analysis-based Deep CNN model. The proposed model is analyzed and validated with the
existing techniques. To prove the performance of the proposed DCNN-G-HHO is highly effective
than existing. Therefore, the performance analysis parameters such as accuracy, recall, precision,
f1-measure are considered. Furthermore, the datasets are initially divided into training and testing,
where 70% of datasets are utilized for training, and the remaining 30% are being used for testing.
The implementation of the suggested DCNN-G-HHO is done through the Python platform. The
Python source code of DCNN-G-HHO can be accessible in [15].
The following sub-sections are explanations on data description, Implementation compo-
nents, and performance evaluation.

5.1 Dataset description

The brain MRI images are obtained from the open-source datasets from the internet Kaggle
[7]. In this work, the MRI scanned image-based datasets are downloaded and utilized. The
obtained datasets are fed into Deep CNN-IWOA. That mainly focuses on detecting brain
tumor detection and classification by using the deep learning technique. Therefore, MRI-based
brain tumor classification can be obtained from massive datasets. The total number of input
images is utilized before augmentation is 248, and once after augmentation is completed, the
input images are turned into 2073. The sample input images of the proposed work are given in
Fig. 3. Also, Fig. 4 shows the brain tumor image.

Fig. 3 Sample input images

Multimedia Tools and Applications (2022) 81:44059–44086 44077

Fig. 4 The brain tumor detected image

The following Fig. 4 shows the sample images of brain tumor affected MRI images. In each
MRI brain image, a strange shape contaminated is a brain tumor. Figure 5 shows the
augmentation samples of the proposed work.
The other section describes the implementation requirements of the proposed model.

5.2 Implementation components

In this part, the experimental results are discussed and clarified. By using the python software,
the implementation process is executed. The performance analysis is conducted to evaluate the
proposed approach with the existing precision, F-measure, accuracy, and recall methods. For
the training process, 70% of the dataset is used, and for the testing process, 30% of the data set is
used for executing the process. The execution process is performed on Windows 10 operating

Fig. 5 Augmented images

44078 Multimedia Tools and Applications (2022) 81:44059–44086

system with 16 GB RAM. The epoch values are taken from 0 to 500. According to [32], the best
accuracy is achieved by the 1024 batch size. The author states that based on their results, the
higher the batch size, the higher the accuracy, meaning that the batch size has a massive impact
on the CNN performance. The batch size is set to 1024 for each epoch on this premise. Also, the
kernel size is considered at 5 X 5. Additionally, our proposed classifier is evaluated against the
state-of-the-art algorithms regarding execution time and memory usage on the MRI dataset.

6 Evaluation and result analysis

This section is divided into three sub-sections: performance evaluation, comparative perfor-
mance analysis, and statistical performance analysis.

6.1 Performance evaluation

The presented approach can be estimated by some statistical measurements, such as true
negative (TN), true positive (TP), False Negative (FN), and False positive (FP) [44]. These
values are represented as a confusion matrix in Table 3.
By using the following equations, these statistical parameter values are evaluated, and those
equations are formulated below.

6.1.1 Accuracy

Accuracy refers to the ratio of the true patterns to the summation of entire patterns. It can be
expressed as
TN þ TP 929 þ 1075
Accuracy ¼ ¼ ¼ 0:97
TN þ TP þ FN þ FP dx2065
Therefore, the accuracy for the proposed DCNN-G-HHO is 0.97.

6.1.2 Precision

Precision can be described as the ratio of true positive to the summation of entire positive
patterns. It can be expressed as

Table 3 Confusion matrix to describe statistical values

n=2065 Output

Perfect Defected
NO YES

Perfect TP=1075 FP=10 1085

NO
Defected FN=51 TN=929 980
YES
1126 939
Multimedia Tools and Applications (2022) 81:44059–44086 44079

TP 1075
Precision ¼ ¼ ¼ 0:99
TP þ FP 1075 þ 10

Thus, precision for the introduced DCNN-G-HHO is 0.99

6.1.3 Recall

This recall parameter is used for calculating the false positive and true positive values, and also
it can be called sensitivity. It can be expressed as
TP 1075
Recall ¼ ¼ ¼ 0:95
TP þ FN 1075 þ 51
The recall value for the suggested DCNN-G-HHO is 0.95.

6.1.4 F-measure

To measure the value of F-measure, it requires both precision and recall values. It can be
expressed as
Precision  Recall 0:99  0:954
Fmeasure ¼ 2  ¼2 ¼ 0:97
Precision þ Recall 0:99 þ 0:954
The F-measure value for the proposed approach is 0.97.
Figure 6 shows that the x-axis represents the false positive rate, and the y- axis represents
the true positive rate.
Figure 7 shows that the x-axis represents the epoch value, and the y-axis represents the loss
value. The blue color represents the training process, and the orange color represents the
testing process. The epoch values are taken from 0 to 500.
In Fig. 8, the x-axis represents the epoch, and the y-axis represents the accuracy. The epoch
values are taken from 0 to 500. The blue color represents the training process, and the orange
color represents the testing process.

Fig. 6 ROC curve for the proposed model

44080 Multimedia Tools and Applications (2022) 81:44059–44086

Fig. 7 The graphical illustration of Loss

Fig. 8 Accuracy for the proposed model

Multimedia Tools and Applications (2022) 81:44059–44086 44081

6.2 Comparative performance analysis

In this part, the comparative analysis is described for the presented approach and the
existing approaches such, as (Long Short-Term Memory (LSTM), Artificial Neural Net-
work (ANN), CNN, SVM, Wide & Deep Learning, Deep Factorization Machines
(DeepFM), SR-FCM-CNN, and Whale Harris Hawks optimization with DeepCNN
(WHHO-based DeepCNN)). The proposed and current methods values are given in
Table 4.
The above table denotes the results of the existing and presented approach. The accuracy
obtained for the brain tumor detection from our approach is 0.97. The accuracy of the
competitive algorithms, such as LSTM, ANN, CNN, SVM, Wide & Deep Learning,
DeepFM, SR-FCM-CNN, and WHHO-based DeepCNN are 0.91, 0.87, 0.94, 0.90, 0.93,
0.88, 0.89, and 0.96, respectively. The precision level for the presented approach is 0.99,
where for the existing techniques, LSTM, ANN, CNN, SVM, Wide & Deep Learning,
DeepFM, SR-FCM-CNN, and WHHO-based DeepCNN are 0.93, 0.86, 0.94, 0.89, 0.90,
0.92, 0.95, and 0.94, respectively. This shows that the precision level for the presented
approach was high and performed well with an effective result. The recall value for the
suggested method is 0.94, where for the existing algorithms, LSTM, ANN, CNN, SVM,
Wide & Deep Learning, DeepFM, SR-FCM-CNN, and WHHO-based DeepCNN are 0.88,
0.85, 0.89, 0.86, 0.87, 0.87, 0.90, and 0.91, respectively. At last, the F-measure value for
the presented method is 0.96, where for the existing methods, LSTM, ANN, CNN, SVM,
Wide & Deep Learning, DeepFM, SR-FCM-CNN, and WHHO-based DeepCNN are 0.92,
0.86, 0.93, 0.88, 0.92, 0.89, 0.94, and 0.93, respectively. Therefore, the presented approach
is more efficient in accuracy, precision, recall, and F-measure from this comparative
analysis than the existing approaches. The comparative analysis graph is represented in
the bar diagram shown in Fig. 9.
The x-axis represents brain tumor detection performance metrics in the above Figure, such
as accuracy, precision, recall, and F-measure. The y-axis represents the parameter values. The
presented approach is denoted by green color. LSTM, ANN, CNN, SVM, Wide & Deep
Learning, DeepFM, SR-FCM-CNN, and WHHO-based DeepCNN, are indicated by golden,
dark gray, brown, dark blue, green, light blue, yellow, light gray, and orange. This coloring
clearly shows that the presented approach is performed efficiently in accuracy, precision,
recall, and f-measure compared with the existing algorithms. Thus, the proposed method is
most preferred for detecting brain tumors.

Table 4 Values for the comparative analysis

Classification technique Accuracy precision Recall F-measure

LSTM 0.91 0.93 0.88 0.92

ANN 0.87 0.86 0.85 0.86
CNN 0.94 0.94 0.89 0.93
SVM 0.90 0.89 0.86 0.88
Wide & Deep Learning 0.93 0.90 0.87 0.92
DeepFM 0.88 0.92 0.87 0.89
SR-FCM-CNN 0.89 0.95 0.90 0.94
WHHO-based DeepCNN 0.96 0.94 0.91 0.93
The proposed DCNN-G-HHO 0.97 0.99 0.95 0.97
44082 Multimedia Tools and Applications (2022) 81:44059–44086

1.05

0.95

0.9

0.85

0.8

0.75
F-measure Recall precision Accuracy

LSTM ANN CNN

SVM Wide & Deep Learning DeepFM
SR-FCM-CNN WHHO-based DeepCNN The proposed DCNN-G-HHO

Fig. 9 Bar diagram for the comparative analysis

6.3 Statistical performance analysis

This section analyses the overall performance benchmarking of the algorithms (execution
time/memory consumption) based on the augmented MRI image dataset. Table 5 presents the
execution time with memory consumption for the 30 solutions obtained by the proposed
DCNN-G-HHO and its nine state-of-the-art algorithms. We observe that our suggested model
exhibited a shorter execution time for detecting and classifying the brain tumors in the MR
images. Similarly, our proposed method consumed less memory than its counterpart tech-
niques. Surprisingly, KNN had a slower execution time than the other techniques, whereas
SVM required the highest memory allocation than its competitive algorithms. Generally, the
suggested DCNN-G-HHO technique had a faster execution and consumed less memory than
the other classifiers.

7 Conclusion

The main intention of the proposed DCNN-G-HHO was to provide an automated detection
model for brain tumors. Therefore, a deep learning-based CNN was considered and enhanced

Table 5 Average execution time with memory consumption for the 30 solutions obtained by the proposed
DCNN-G-HHO against its counterpart algorithms

Algorithms Execution time Memory consumption (MB)

LSTM 26.122 105.109

ANN 37.128 102.085
KNN 41.013 95.101
Deep KNN 38.891 97.982
SVM 40.172 103.138
Wide & Deep Learning 39.183 101.283
DeepFM 30.837 98.012
SR-FCM-CNN 22.193 79.373
WHHO-based DeepCNN 23.148 89.172
The proposed DCNN-G-HHO 18.091 74.183
Multimedia Tools and Applications (2022) 81:44059–44086 44083

by adopting a hybrid optimization algorithm with a combination of GWO and HHO tech-
niques. Moreover, segmentation plays a significant role in detecting tumors. Accordingly, an
Otsu thresholding technique is utilized. As a result, effective segmentation and classification
were achieved better results.
The technique’s performance was ensured by comparing it with the existing technique on
huge augmented MRI images in terms of accuracy, precision, recall, f-measure, execution
time, and memory consumption. At first glance, the performance comparison proved that the
suggested DCNN-G-HHO was highly effective than current techniques. With 0.97 accuracies
for brain tumor diagnosis, the recommended DCNN-G-HHO method outperformed current
approaches. This demonstrates the approach’s accuracy and effectiveness. The proposed
approach has a recall of 0.95, whereas the existing methods had lower recollections than our
method. Finally, the F-measure for the proposed technique is 0.96. Therefore, the proposed
methodology is more efficient than the current methods regarding accuracy, precision, recall,
and F-measure. Additionally, the overall performance of our proposed approach was
benchmarked in terms of execution time and memory use on the enhanced MRI image dataset.
The suggested DCNN-G-HHO was faster at identifying and categorizing brain tumor cancers
on the MR images. Similarly, our solution used less RAM than its competing methods.
Surprisingly, KNN took longer than the other methods, but SVM consumed the most memory.
The DCNN-G-HHO approach was quicker and used less memory than the different classifiers.
The main contributions of the newly introduced DCNN-G-HHO were:

& High accuracy and precision of detecting and classifying brain tumors due to the use of G-
HHO and Otsu thresholding combined with DCNN;
& Massive augmented MRI image datasets were used to provide reliable results;
& The performance of the suggested DCNN-G-HHO is improved and enabled better detec-
tion and classification results.

Thus, we have designed DCNN-G-HHO as a combination of DCNN and G-HHO with few
exploratory and exploitative mechanisms. It is also viable to use other evolutionary strategies
such as mutation and crossover, multi-swarm and multi-leader structures, evolutionary
updating structures, and chaos-based phases. These operators and concepts are advantageous
for future works. Binary and multi-objective variants of G-HHO may be developed in future
developments. Additionally, it may be used to address a variety of issues in engineering and
other industries. Another intriguing avenue is to examine various ways of coping with
constraints in real-world restricted scenarios.
Because of its high accuracy, DCNN-G-HHO could be deployed on different applications.
Besides, the modifications, as well as hybridizations of the algorithm, have been a hot spot for
scholars. But still, there is room for others to work in improving our introduced algorithm as it
can be hybridized with most recent swarm algorithms, such as backtracking search optimiza-
tion algorithm [19, 20], the variants of evolutionary clustering algorithm star [21, 24–26],
chaotic sine cosine firefly algorithm [13], and hybrid artificial intelligence algorithms [22].
Furthermore, DCNN-G-HHO can be applied to more complex and real-world applications to
explore more deeply the advantages and drawbacks of the algorithm or improve its efficien-
cies, such as engineering application problems [13], laboratory management [49], e-
organization and e-government services [23], online analytical processing [16], web science
[14], the Semantic Web ontology learning [17], cloud computing paradigms [34–37], and
evolutionary machine learning techniques [18, 26, 46].
44084 Multimedia Tools and Applications (2022) 81:44059–44086

Acknowledgments The authors would like to express their sincere gratitude to their respective institutions for
providing the facilities and ongoing assistance necessary to perform this research.

Declarations

Conflict of interest There is no conflict of interest.

References

1. Akagic A, Buza E, Omanovic S, Karabegovic A (2018) Pavement crack detection using Otsu thresholding
for image segmentation. In: 2018 41st international convention on information and communication
technology, Electronics and Microelectronics (MIPRO). IEEE, pp. 1092–1097
2. Al-Tashi Q, Rais HM, Abdulkadir SJ, et al (2020) A review of grey wolf optimizer-based feature selection
methods for classification. Evol Mach Learn Tech 273–286
3. Amin J, Sharif M, Raza M, Saba T, Anjum MA (2019) Brain tumor detection using statistical and machine
learning method. Comput Methods Prog Biomed 177:69–79
4. Amin J, Sharif M, Yasmin M, Fernandes SL (2020) A distinctive approach in brain tumor detection and
classification using MRI. Pattern Recogn Lett 139:118–127
5. Anaraki AK, Ayati M, Kazemi F (2019) Magnetic resonance imaging-based brain tumor grades classifica-
tion and grading via convolutional neural networks and genetic algorithms. Biocybern Biomed Eng 39:63–
74
6. Arasi PRE, Suganthi M (2019) A clinical support system for brain tumor classification using soft computing
techniques. J Med Syst 43:1–11
7. Chakrabarty N (2019) Brain MRI images for brain tumor detection. In: Kaggle. https://www.kaggle.com/
navoneel/brain-mri-images-for-brain-tumor-detection. Accessed 7 Apr 2021
8. Chen S, Ding C, Liu M (2019) Dual-force convolutional neural networks for accurate brain tumor
segmentation. Pattern Recogn 88:90–100
9. Deepak S, Ameer PM (2019) Brain tumor classification using deep CNN features via transfer learning.
Comput Biol Med 111:103345
10. Faris H, Aljarah I, Al-Betar MA, Mirjalili S (2018) Grey wolf optimizer: a review of recent variants and
applications. Neural Comput Appl 30:413–435
11. Ghassemi N, Shoeibi A, Rouhani M (2020) Deep neural network with generative adversarial networks pre-
training for brain tumor classification based on MR images. Biomed Signal Process Control 57:101678
12. Goceri E (2020) CapsNet topology to classify tumours from brain images and comparative evaluation. IET
Image Process 14:882–889
13. Hassan BA (2020) CSCF: a chaotic sine cosine firefly algorithm for practical application problems. Neural
Comput Appl 1–20
14. Hassan BA (2021) Analysis for the overwhelming success of the web compared to microcosm and hyper-G
systems. arXiv Prepr arXiv210508057
15. Hassan BA (2022) DCNN-G-HHO. https://github.com/bryarahassan/DCNN-G-HHO.git. Accessed 20
Jan 2022
16. Hassan BA, Qader SM (n.d.) A new framework to adopt multidimensional databases for organizational
information Sys-tem Strategies
17. Hassan B, Dasmahapatra S (n.d.) Towards Semantic Web: Challenges and Needs
18. Hassan BA, R TA (2020) Formal context reduction in deriving concept hierarchies from corpora using
adaptive evolutionary clustering algorithm. Complex Intell Syst
19. Hassan BA, Rashid TA (2019) Operational framework for recent advances in backtracking search optimi-
sation algorithm: a systematic review and performance evaluation. Appl Math Comput 124919
20. Hassan BA, Rashid TA (2020) Datasets on statistical analysis and performance evaluation of backtracking
search optimisation algorithm compared with its counterpart algorithms. Data Br 28:105046
21. Hassan BA, Rashid TA (2021) A multidisciplinary ensemble algorithm for clustering heterogeneous
datasets. Neural Comput Appl https://doi.org/10.1007/s00521-020-05649-1
22. Hassan BA, Rashid TA (2021) Artificial intelligence algorithms for natural language processing and the
semantic web ontology learning. arXiv Prepr arXiv210813772
23. Hassan BA, Ahmed AM, Saeed SA, Saeed AA (2016) Evaluating e-government Services in Kurdistan Institution
for strategic studies and scientific research using the EGOVSAT model. Kurdistan J Appl Res 1:1–7
Multimedia Tools and Applications (2022) 81:44059–44086 44085

24. Hassan BA, Rashid TA, Hamarashid HK (2021) A novel cluster detection of COVID-19 patients and
medical disease conditions using improved evolutionary clustering algorithm star. Comput Biol Med 104866
25. Hassan BA, Rashid TA, Mirjalili S (2021) Formal context reduction in deriving concept hierarchies from
corpora using adaptive evolutionary clustering algorithm star. Complex Intell Syst 7:1–16
26. Hassan BA, Rashid TA, Mirjalili S (2021) Performance evaluation results of evolutionary clustering
algorithm star for clustering heterogeneous datasets. Data Br 107044
27. Heidari AA, Aljarah I, Faris H, et al (2019) An enhanced associative learning-based exploratory whale
optimizer for global optimization. Neural Comput Appl 1–27
28. Heidari AA, Mirjalili S, Faris H, Aljarah I, Mafarja M, Chen H (2019) Harris hawks optimization: algorithm
and applications. Futur Gener Comput Syst 97:849–872
29. Huang P, Ye Q, Zhang F, Yang G, Zhu W, Yang Z (2021) Double L2, p-norm based PCA for feature
extraction. Inf Sci (Ny) 573:345–359
30. Islam MK, Ali MS, Miah MS, et al (2021) Brain tumor detection in MR image using superpixels, principal
component analysis and template based K-means clustering algorithm. Mach learn with Appl 100044
31. Kaldera H, Gunasekara SR, Dissanayake MB (2019) Brain tumor classification and segmentation using
faster R-CNN. In: 2019 advances in science and engineering technology international conferences (ASET).
IEEE, pp 1–6
32. Kandel I, Castelli M (2020) The effect of batch size on the generalizability of the convolutional neural
networks on a histopathology dataset. ICT express 6:312–315
33. Krishnakumar S, Manivannan K (2021) Effective segmentation and classification of brain tumor using
rough K means algorithm and multi kernel SVM in MR images. J Ambient Intell Humaniz Comput 12:
6751–6760
34. Kumar M, Sharma SC (2018) Deadline constrained based dynamic load balancing algorithm with elasticity
in cloud environment. Comput Electr Eng 69:395–411
35. Kumar M, Sharma SC, Goel A, Singh SP (2019) A comprehensive survey for scheduling techniques in
cloud computing. J Netw Comput Appl 143:1–33
36. Kumar M, Dubey K, Pandey R (2021) Evolution of emerging computing paradigm cloud to fog: applica-
tions, limitations and research challenges. In: 2021 11th international conference on cloud computing, Data
Science & Engineering (Confluence). IEEE, pp. 257–261
37. Kumar M, Kishor A, Abawajy J et al (2021) ARPS: An autonomic resource provisioning and scheduling
framework for cloud platforms. IEEE Trans Sustain Comput
38. Kurup RV, Sowmya V, Soman KP (2019) Effect of data pre-processing on brain tumor classification using
capsulenet. In: International Conference on Intelligent Computing and Communication Technologies.
Springer, pp. 110–119
39. Marghalani BF, Arif M (2019) Automatic classification of brain tumor and Alzheimer's disease in MRI.
Procedia Comput Sci 163:78–84
40. Muhammad K, Khan S, Del Ser J, de Albuquerque VHC (2020) Deep learning for multigrade brain tumor
classification in smart healthcare systems: a prospective survey. IEEE Trans Neural Networks Learn Syst
32:507–522
41. Ostrom QT, Adel Fahmideh M, Cote DJ et al (2019) Risk factors for childhood and adult primary brain
tumors. Neuro-Oncology 21:1357–1375
42. Özyurt F, Sert E, Avcı D (2020) An expert system for brain tumor detection: fuzzy C-means with super
resolution and convolutional neural network with extreme learning machine. Med Hypotheses 134:109433
43. Pandiselvi T, Maheswaran R (2019) Efficient framework for identifying, locating, detecting and classifying
MRI brain tumor in MRI images. J Med Syst 43:1–14
44. Powers DMW (2020) Evaluation: from precision, recall and F-measure to ROC, informedness, markedness
and correlation. arXiv Prepr arXiv201016061
45. Rammurthy D, Mahesh PK (2020) Whale Harris hawks optimization based deep learning classifier for brain
tumor detection using MRI images. J King Saud Univ Inf Sci
46. Rashid TA, H BA (2020) A multi-disciplinary ensemble algorithm for clustering heterogeneous datasets.
Neural Comput Appl
47. Rawat W, Wang Z (2017) Deep convolutional neural networks for image classification: a comprehensive
review. Neural Comput 29:2352–2449
48. Saba T, Mohamed AS, El-Affendi M et al (2020) Brain tumor detection using fusion of hand crafted and
deep learning features. Cogn Syst Res 59:221–230
49. Saeed MHR, Hassan BA, Qader SM (2017) An optimized framework to adopt computer laboratory
administrations for operating system and application installations. Kurdistan J Appl Res 2:92–97
50. Sahoo DK, Mishra S, Mohanty MN (2020) Wavelet transform and WCA based deep convolutional network
for brain tumor detection and classification from magnetic resonance images. PalArch's J Archaeol Egypt/
Egyptology 17:2319–2328
44086 Multimedia Tools and Applications (2022) 81:44059–44086

51. Sajjad M, Khan S, Muhammad K, Wu W, Ullah A, Baik SW (2019) Multi-grade brain tumor classification
using deep CNN with extensive data augmentation. J Comput Sci 30:174–182
52. Sert E, Özyurt F, Doğantekin A (2019) A new approach for brain tumor diagnosis system: single image
super resolution based maximum fuzzy entropy segmentation and convolutional neural network. Med
Hypotheses 133:109413
53. Sharif M, Amin J, Raza M, Yasmin M, Satapathy SC (2020) An integrated design of particle swarm
optimization (PSO) with fusion of features for detection of brain tumor. Pattern Recogn Lett 129:150–157
54. Sultan HH, Salem NM, Al-Atabany W (2019) Multi-classification of brain tumor images using deep neural
network. IEEE Access 7:69215–69225
55. Talo M, Yildirim O, Baloglu UB, Aydin G, Acharya UR (2019) Convolutional neural networks for multi-
class brain disease detection using MRI images. Comput Med Imaging Graph 78:101673
56. Tang C, Liu X, Zhu X, Xiong J, Li M, Xia J, Wang X, Wang L (2019) Feature selective projection with
low-rank embedding and dual Laplacian regularization. IEEE Trans Knowl Data Eng 32:1747–1760
57. Tang C, Zheng X, Liu X, Zhang W, Zhang J, Xiong J, Wang L (2021) Cross-view locality preserved diversity
and consensus learning for multi-view unsupervised feature selection. IEEE Trans Knowl Data Eng 1
58. Thaha MM, Kumar KPM, Murugan BS, Dhanasekeran S, Vijayakarthick P, Selvi AS (2019) Brain tumor
segmentation using convolutional neural networks in MRI images. J Med Syst 43:1–10
59. Toğaçar M, Ergen B, Cömert Z (2020) BrainMRNet: brain tumor detection using magnetic resonance
images with a novel convolutional neural network model. Med Hypotheses 134:109531
60. Vijh S, Sharma S, Gaurav P (2020) Brain tumor segmentation using OTSU embedded adaptive particle
swarm optimization method and convolutional neural network. In: Data visualization and knowledge
engineering. Springer, pp. 171–194
61. Yin B, Wang C, Abza F (2020) New brain tumor classification method based on an improved version of
whale optimization algorithm. Biomed Signal Process Control 56:101728
62. Zeineldin RA, Karar ME, Coburger J, Wirtz CR, Burgert O (2020) DeepSeg: deep neural network
framework for automatic brain tumor segmentation using magnetic resonance FLAIR images. Int J
Comput Assist Radiol Surg 15:909–920

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.