Aggarwal et al.

BMC Medical Informatics and

BMC Medical Informatics and Decision Making (2023) 23:78
https://doi.org/10.1186/s12911-023-02174-8 Decision Making

RESEARCH Open Access

An early detection and segmentation

of Brain Tumor using Deep Neural Network
Mukul Aggarwal1, Amod Kumar Tiwari2, M Partha Sarathi3 and Anchit Bijalwan4*

Abstract
Background Magnetic resonance image (MRI) brain tumor segmentation is crucial and important in the medical
field, which can help in diagnosis and prognosis, overall growth predictions, Tumor density measures, and care plans
needed for patients. The difficulty in segmenting brain Tumors is primarily because of the wide range of structures,
shapes, frequency, position, and visual appeal of Tumors, like intensity, contrast, and visual variation. With recent
advancements in Deep Neural Networks (DNN) for image classification tasks, intelligent medical image segmentation
is an exciting direction for Brain Tumor research. DNN requires a lot of time & processing capabilities to train because
of only some gradient diffusion difficulty and its complication.
Methods To overcome the gradient issue of DNN, this research work provides an efficient method for brain Tumor
segmentation based on the Improved Residual Network (ResNet). Existing ResNet can be improved by maintaining
the details of all the available connection links or by improving projection shortcuts. These details are fed to later
phases, due to which improved ResNet achieves higher precision and can speed up the learning process.
Results The proposed improved Resnet address all three main components of existing ResNet: the flow of informa-
tion through the network layers, the residual building block, and the projection shortcut. This approach minimizes
computational costs and speeds up the process.
Conclusion An experimental analysis of the BRATS 2020 MRI sample data reveals that the proposed methodology
achieves competitive performance over the traditional methods like CNN and Fully Convolution Neural Network
(FCN) in more than 10% improved accuracy, recall, and f-measure.
Keywords Brain tumor, Segmentation, ResNet, Deep neural network, CNN, Healthcare, Prediction models

Introduction deep-learning network architectures. The processing of

Brain Tumor segmentation and detection are very chal- medical images plays a crucial role in assisting humans
lenging in the medical imaging area. Various DNN meth- in identifying different diseases [1]. Classification of brain
ods are used for Tumor segmentation, utilizing multiple Tumors is a significant part that depends on the expertise
and knowledge of the physician. An intelligent system for
detecting and classifying brain Tumors is essential to help
*Correspondence: physicians. Gliomas have an irregular shape and ambigu-
Anchit Bijalwan ous boundaries, which are the most challenging Tumors
[email protected]
1
Dr. A.P.J. Abdul Kalam Technical University, Lucknow, Uttar Pradesh, India to detect. Various authors have performed additional
2
Rajkiya Engineering College, Sonbhadra, Uttar Pradesh, India research on deep learning networks based on healthcare,
3
Amity School of Engineering and Technology, Amity University, Noida, i.e., Convolutional neural networks (CNNs), LinkNet,
Uttar Pradesh, India
4
Faculty of Electrical and Computer Engineering, Arba Minch University, Visual Graphic Group (VGG), UNet, and SegNet [2].
Arba Minch, Ethiopia Image segmentation poses significant challenges,
including categorization, image processing, object

© The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. The Creative Commons Public Domain Dedication waiver (http://​creat​iveco​
mmons.​org/​publi​cdoma​in/​zero/1.​0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 2 of 12

recognition, and explanation. Whenever an image clas- • The simple transformation used in Enhanced ResNet
sification model is formed, e.g., it must be eligible to mainly improves the training process of Convolu-
function with great precision even when subjected to tional models by utilizing the “shortcut links.” These
occlusion, lighting modifications, observing angles, and links provide all the possible route details in a single
other factors [3]. place and provide access in a single click reducing the
The conventional object detection process, includ- accessing time.
ing its primary feature extraction step, is unsuitable for
wealthy areas. Sometimes experts in the domain cannot The complete research article is organized as follows:
provide a single or collective of functionalities capable Section 1 covers the introduction, Section 2 covers exist-
of achieving accurate results under varying conditions. ing Tumor segmentation work related to research, Sec-
The concept of model training emerges due to that kind tion 3 covers material and methods, section 4 covers
of problem. The appropriate features for working with results, section 5 covers the discussion and Section 6 cov-
image data are instantly figured out [4]. ers the conclusion and future direction of the research.
Content-based image retrieval provides various imag-
ing modalities, such as CT, MR, PET, X-rays, and Ultra- Related works
sound. Also, the many image data available because of The field of Tumor segmentation is continuously under-
different scan parameter settings and multiple views of going investigation. Deep learning has recently proven
the same pathology make image retrieval in the medical effective in healthcare image segmentation and informa-
domain tough and challenging. However, at the same tion extraction. In deep learning techniques, pixel-based
time, it is one of the essential applications [5]. The MR classification is the latest phenomenon. Various research-
images are taken from three different directions. These ers have suggested different methods for brain Tumor
views are called sagittal, axial, and coronal [6]. For segmentation. This section covers the analysis of a few of
CBIR to be used in healthcare as a diagnostic aid, the the critical research.
medical information framework must be robust in vari- Research [9] presents brain Tumor segmentation using
ous scenarios to be accepted by clinicians and medical DNN. Brain Tumors are segmented on magnetic reso-
practitioners [7]. nance visuals of the brain using a Deep Convolutional
First, case-based reasoning will be more acceptable encoder model. This approach enhances learning by
to the medical community when the retrieval engine extracting attributes from complete images, eliminating
results in cases with exact locations and similar pathol- patchwork selections, and improving calculations at adja-
ogy responding to a query (new) case [8]. cent intersections. Research [10] presented a technique
This will significantly help the medical expert have for the early detection of brain cancers. Magnetic reso-
more information about the case and aid the expert nance images were examined to identify Tumor-bearing
in monitoring. Secondly, the database formed for test- areas and categorize them into various classifications. In
ing purposes should be carefully built consisting of image classification techniques, deep learning generates
cases from multiple views, different scanning param- efficient performance.
eters, and acquired from different imaging modalities. Consequently, the Fully Convolutional Networks tech-
CNN has been used to segment Tumors in multi-modal nique was applied and incorporated through the Tensor
Imaging [8]. Flow repository throughout this research. A newer CNN
The CNN architecture is sophisticated, combining technique has been demonstrated to have a precision of
segmentation and classification into a single product. 91 percent, which is better than previous research.
Current segmentation methods have been designed to Research [11] developed a model by utilizing Brain
solve the reduplication issue of CNNs by allocating a imaging to recognize the nature of brain Tumors. A two-
target class toward each pixel. A CNN model has been dimensional CNN was used to acknowledge malignant
transformed into an FCN (Fully CNN). This article has Tumors with an accuracy rate of 93 percent. The data for
critical contributions to brain Tumor research, which the four most often detected brain Tumors are included
are as follows: in the research’s analysis.
Research [12] advised a responsive and efficient Tumor
• This research develops the ResNet Model to address segmentation framework. In a Cascades Classifica-
the weaknesses of CNN and FCN methodologies tion Model, this strategy reduces computation time and
and improve computational costs. The principle of addresses the problem of overfitting. Using two sepa-
ResNet is premised on adding the layer’s outcome rate forms, this CNN architecture extracts global and
towards its significant input. regional characteristics. Additionally, the Tumor detec-
tion precision is significantly enhanced compared to
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 3 of 12

Fig. 1 Architecture of Convolution Neural Network (CNN)

current algorithms. The average WT, increasing Tumor, The MLP-IWOA-based classification was used to classify
and Tumor center dice scores for the proposed approach Tumors and outperformed specific current approaches.
achieved 92.3%, 94.5%, and 93.2 %. Research [18] consolidated significant statistical attrib-
Research [13] developed a model to evaluate Tumors utes with CNN architectures to create a technique for the
utilizing an MRI dataset. It entails finding cancer, grading segment of brain cancer cells. The architecture concen-
it by size and type, and determining the Tumor’s position. trated on the Tumor’s boundary. The two-dimensional
Instead of using alternative approaches for each classifi- Wavelet Decomposition, Gabor Filters Filter, and similar-
cation task, this strategy used a single model to organize ity measures were used to identify and extract the image.
MRI Images on many classification techniques. A significant feature with further categorization was
Research [14] prompted brain Tumor identification and developed by combining these statistical properties.
separation by integrating both training methods. The first Research [19] analyzed that cancer seems to be the
proposed approach was the Binary Pattern method based most severe disease and therefore is considered challeng-
upon that neighbor range connection termed ‘nLBP’. ing to treat. While behind the bottom section of the belly
The second strategy was based on the perspective of the is a pancreatic malignant that develops in the pancre-
neighbor next door called “αLBP.” The above two tech- atic cells that aid indigestion. Its stage of growth deter-
niques were developed to process and analyses MRI mines the therapy for this Tumor. The Tumor is detected
images of the most prevalent cancers: Glioblastoma, by individually identifying the afflicted region of the CT
malignant Tumors, & gland Tumors. For feature evo- scanned data. It forecasts the Tumor region under con-
lution, the statistics of the precompiled images were sideration by utilizing Gaussian Mixture Framework and
employed. Conventional extraction of feature strategies Expectation-Maximization method & CNN [20].
scored worse than this proposed model.
Research [15] applied the brain Tumor partition by
integrating all the RELM (“Regularized Extreme Learning Materials & Methods
Machine”). The procedure initially normalized images to This section covers the essential methods used in this
make the framework’s understanding easier. The frame- research and the proposed improved ResNet method
work utilized a min-max strategy for pre-processing working.
phase. This min-max processing method significantly
improved the brightness of the original images. Convolution Neural Network
Research [16] applied the brain Tumor partition by CNN is mainly a deep learning approach used to clas-
integrating all the RELM (“Regularized Extreme Learning sify images. CNN is an artificial neural network designed
Machine”). The procedure initially normalized images to to analyze input in a mesh form. In CNN, a Convolu-
make the framework’s understanding easier. The frame- tion process is an activity inside the convolution layer
work utilized a min-max strategy for pre-processing premised on just a mathematical matrix operation that
phase. This min-max processing method significantly increases the matrix of both the filtration system in the
improved the brightness of the original images. image to be analyzed. This convolution operation is the
Research [17] proposed a Convolutional Percep- first and most significant utilization phase [21].
tron neural network-based segmentation initiative to Figure 1 shows the architecture of CNN. This figure
improve the Whale Optimization method. For improved shows three layers named convolutional, pooling and
feature evolution and partition, the hybrid algorithm fully connected layers. Another layer often employed is
produced an updated form of WOA. The Mean Filtering a pooling layer that receives the whole or averaged values
was used to first remove the noise from data in product of the pixels image regions. CNN is capable of learning
development and production. The enhanced WOA was advanced functionality by creating a feature map.
used to pick characteristics from the retrieved features.
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 4 of 12

Fig. 2 FCN Architecture

It constructs many feature maps; each convolution is linked to every cell inside the subsequent stage. Step
layer core is covered across its input sequence. Input 6: At the end of the learning process, a failure layer is
sequences recognize characteristics presented on this applied to provide constructive feedback on the CNN
feature map as simple boxes. Such maps are sent to architecture.
the optimum related resources layer, keeping the most
important features while discarding the remaining. Inside Fully Convolutional Network (FCN)
each fully-connected layer, the characteristics of its max- In research [24], the FCN has been suggested as a solu-
pooling base layer are turned into a 1-D feature vector, tion to semantic segmentation and classification.
which will be employed to determine the output conse- Researchers utilized AlexNet, VGGNet, and GoogleNet
quence [22]. Image scalability is not possible in a tradi- as potential options. Researchers transmitted all such
tional neural network model. approaches from classification methods to thick FCN by
However, in a CNN model, the image can be scaled replacing convolution layers with (1×1) Convolutional
(that is, it can go from a 3D input space to a 3-dimen- layers and adding a (1 × 1) convolution to frequency axis
sional output pattern). The CNN Model comprises its 21 to forecast rankings at each class and context category.
input layers, convolution, Rectified Unit layer, pool- FCN can learn to quickly build dense assumptions for
ing layer, and fully-Connected layers. The provided per-pixel processes such as semantic segmentation [24].
data (input images) gets split into small sections inside Figure 2 shows the working of FCN architecture
the convolution operation. The ReLU layer performs for image segmentation. Each layer in FCN is just a
element-by-element activation. The requirement for a 3-D array of different sizes, including height, width,
pooling layer is voluntary. Here the option of using or and dimension. The image is the first layer, with all the
skipping can be taken pixels’ information, including height, width, and col-
On the other hand, this pooling layer is mainly uti- our space dimensions. Higher-level locations correlate
lized for downstream sampling. A category score or class to the image regions and are route-based, their visual
score code is represented in the last stage (i.e., fully con- field.
nected layer) based on 0 and 1. The CNN-based brain Significant alterations in FCN that further contributed
Tumor segmentation training/testing rounds are catego- to the conceptual framework to accomplish state- of-art
rized into two sections. All images are classified using outcomes are just the prototype VGG16, bipolar extrapo-
categories like Tumor images and non-Tumor brain lation method for up-sampling only the resulting feature
Tumor images [23]. outline, and skip correlation for incorporating minimal
Algorithm: 1 CNN-based Brain Tumor segmentation layer as well as consistently high layer characteristics in
process. Input: Brain Tumor imagoes dataset Output: the closing layer for fine-grained segmentation. FCN only
Tumor images are segmented into Tumor and Non- uses local data for segmentation.
Tumor images. Step 1: Impose a Convolutional filtration However, only neighborhood details make logical seg-
to the very initial layer. Step 2: Refine the Convolutional mentation unclear because the image’s global semantic
filter to lower its sensitivities called “sub-sampling.” Step scope is lost. Relevant information first from the entire
3: All signal transmissions from one layer to the next are image is beneficial for reducing uncertainty. U-Net and
regulated primarily through activation blocks. Step 4: V-Net are the most popular FCN architectures widely
Use the rectified linear component to shorten the train- used in image segmentation [25, 26].
ing process. Step 5: Each neuron in the previous layer
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 5 of 12

Proposed model based on Residual Learning Network RL(RX) = H(RX) − RX (2)

The work explains the MRI brain Tumor datasets for
medical image analysis that are freely available. This where H (RX) represents the actual outcome, RL rep-
research outlines the performance indicators for evaluat- resents the Residual learning value, and RX represents
ing deep learning image and segmentation models. the input. To overcome the gradient issue of DNN, this
To address existing challenges, this work utilized an research provides an efficient method for a brain Tumor.
advanced pre-processing approach in the proposed
method to eliminate many irrelevant data, resulting in
impressive outcomes, perhaps in the current convolu- The Proposed Improved ResNet Model Working
tional neural network. Segmentation based on the Improved Residual Learn-
The proposed strategy does not employ a complicated ing Network (ResNet). Existing ResNet can be improved
segmentation method to categorize the position of the by maintaining the details of all the available connection
brain Tumor and the extraction of features, which results links. The proposed ResNet utilizes a jump relationship
in a time-consuming process with a high fault rate. in that initial input data is combined with the convolu-
ResNet has been taken for proposed work as it is free tion building’s outcome. The above addresses the disap-
from gradient issues, originally a problem of various deep pearing gradient problem by enabling an additional route
learning models. The fading gradient problem occurs for the gradient to move across. The proposed method
during the training procedure of a CNN network. As the also utilizes an identification function that allows a more
learning continued, a gradient rule of previous layers low- significant layer to accomplish as delicate as a bottom
ered to nil or zero. A ResNet method can be utilized to level. The proposed model used the pre-processing, Data
address this problem. A gain of the relationship between Segmentation, and post-processing phases [30–32].
these factors residual layer in ResNet is combined with all Figure 4 presents the working of the proposed ResNet
of its direct input to become its next inner layer [27–29]. model. In improved ResNet, the complete process is
Let H(RX) denote a residual mapping to establish a deep divided into four phases
residual block, as shown in Fig. 3. In past research, researchers suggested numerous
ResNet configurations with ResNet-18, ResNet-34,
H(RX) = F(RX) + RX (1) ResNet-50, and ResNet-152 layers. Each layer of just a
Consider a CNNS block with RX as input and the main ResNet consists of several frames or building blocks. The
objective of learning the accurate distribution H (RX). Identification and Convolutional blocks are merged to
The output and the information difference is the “Resid- produce an Improved ResNet structure in such imple-
ual learning value (RL),” as described in equation 2. mentations. This research uses an improved ResNet-50
layered model for segmentation because it has more fab-
ulous depth layers than ResNet-34 and fewer parameters
than other ResNet models, resulting in a quicker training
period. Figure 4 shows the ResNet-50 architectures [33].
0
Lbce = yi ∗ logOi + 1 − yi ∗ log(1 − Oi) (3)
i

2 0 ∗(Oi ∗ yi)

Ldice = − 0 i 0 (4)
i Oi + i yi

where Lbce represents the standard binary entropy loss

and Ldice represents the dice loss mainly occurring during
image segmentation.
The complete process of the proposed Improved
ResNet is as follows:

• Step 1: It contains a two-dimensional Convolution

that has 64 filtrations of (7*7) framings and just a
stride of size (2*2) small-batch Standard, and also the
Fig. 3 ResNet working structure ReLU (activation function) completes the route axis
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 6 of 12

Fig. 4 (A) Long Skip Connection process in ResNet, (B) ResNet Bottleneck Block process, (C) ResNet Basic Block Working, and (D) ResNet Simple
Block Working

uniformity. Finally, a Max Pooling with a frame of Proposed work model description
(2*2) is used. Phase 1
• Step 2: It includes one two-dimensional CNN model The Residual Network with Long Skip Connections is
block with two Identification blocks, each having represented by Phase 1. It contains down-sampling (in
three pairs of filtrations [64, 64, 256] and a stride Figure 4, represented by blue colour), indicating that it
with size (1*1). is a contracting path. Similarly, an up-sampling (in Fig-
• Step 3: It comprises one fully-connected block with ure 4, represented by orange colour) reveals that it is a
three Identification blocks, each with three pairs of rapidly expanding route. During this process, long skip
filtrations [128, 128, 512] to a stride with size (2*2). connections interact with the contracting path to the
• Step 4: It contains one Convolution layer block as growing direction, shown with arrows from left to right
well as five Identification; it also uses three pairs of in Figure 4A.
filtration of size [256, 256, 1024] and blocks size (3*3),
as well as a stride of size (2*2). Phase 2
• Step 5: It comprises one Convolution layer block and Various (1*1) and (3*3) Conv are used; these blocks are
two Identification blocks, each with three pairs of fil- called bottlenecks. BN and ReLU are used in this phase
trations [512, 512, 2048] with just a stride size (2*2). [34–36]. The concept behind Pre-Activation ResNet is
• Step 6: The fully connected layer is also used to to employ BN-ReLU just before a Conv, as shown in Fig-
reduce the direct input toward the number of sub- ure 4B. the Benefits of using these bottleneck blocks are
classes using a “Soft-max reactivation” algorithm, less training time and improved performance. The use
after which the outcome is flattened. of a bottleneck reduces the number of parameters and
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 7 of 12

matrix multiplications. For example, if 9 operations were MRI images for testing. Figure 5 shows the Brain Tumor
there, it would mainly reduce them to 6. The idea is to types available in the BraTS 2020 dataset.
make residual blocks as thin as possible to increase the
depth and has fewer parameters. Performance measuring parameters
The following essential version was utilized to measure
Phase 3 the performance of the proposed method and the exist-
The third phase is the primary block phase, mainly utiliz- ing one [39–41].
ing (3*3) blocks only, not the (1*1) block. This phase rep-
resents the basic block. A basic ResNet block comprises Mean Square Error (MSE)
two layers of 3x3 conv /BatchNorm/relu. In the picture, The procedure of squaring predicted quantities is MSE.
the lines represent the residual operation. The dotted line An average of such squared errors can be used to explain
means that the shortcut was applied to match the input it. Equation 5 denotes the cumulative square estimation
and the output dimension error between the actual picture and the output image as
MSE
Phase 4
The last phase is the simple block phase, which utilizes
(3*3) n blocks. Max Pooling is used in this phase which 1
m−1
    n−1
[l i, j − K i, j ]}2
 
rejects a big chunk of data. It extracts only the most sali- MSE =
MN
∗{ ∗ (5)
ent features of the data. MaxPool bound the system to i=0 j=0
only the very important features and might miss out on
some details
Peak Signal Noise Ratio (PSNR)
Dataset description
PSNR relates to a picture’s immune function to noise
This research utilized the BraTS2020 dataset [37]. A external interference signals. When the PSNR level is
brat consistently evaluates cutting-edge brain Tumor greater, the noisy interference signal’s effect on the MR
segmentation approaches in composite MRI scan data. image database is minimal. MSE phrases are used to rep-
BraTS 2020 uses multi-institutional like pre Image data. resent PSNR. PSNR must be between 40 and 60 dB. It is
It concentrates on segmenting inherently heterogeneous calculated by Eq. 6. Where Maxl is usually 255 and MSE
(through shape, location, and cell biology) brain Tumors, is the mean square error
such as gliomas. It includes 369 brain Tumor MR images. Max1
As described in Fig. 5, all previous research exam- PSNR = 10log10 (6)
MSE
ined T1-weighted (called T1), post-contrast T1-weighted
(called T1ce), T2-weighted (called T2), and fluid-atten-
uated inversion recovery (called Flair) sequencing. Each Computation Time
of the images has a (240*240*155) size[38]. The dataset is The time it takes to complete the segmentation proce-
collected from the online Kaggle website. It includes 369 dure is calculated in milliseconds or Seconds and repre-
brain MR images; 125 are utilized for training and 169 sented as elapsed time.

Fig. 5 Brain Tumor Images in BraTS2020 (1) for Type T1, (2) for Tumor Type T2, (3) for Tumor Type T1c, and (4) for Tumor type FLAIR
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 8 of 12

Jaccard Coefficient (JC) image sequences of a particular patient, which helps in

It also serves as a metric for evaluating segmentation Tumor growth and sequencing analysis.
strategies. Jacquard offers Eq. 7 to compute the match- This work has presented an improved Recurrent neural
ing of two Q1 and Q2 pairs by standardizing the vol- network-based approach for Tumor segmentation from
ume of their overlap over the respective union. multi-modal 3-dimensional MRI images that further
 utilizes the BraTS 2020 brain Tumor dataset for perfor-
|Q1 Q2|
JC = 2 ∗ (7) mance validation. Several possible solutions have been
|Q1| + |Q2| tried while messing with CNN models. Table 1 shows the
proposed improved ResNet system parameters utilized
for training purposes. After normalization, the Stochastic
Dice Similarity Coefficient (DSC) Gradient Descent optimization method (SGDOM) man-
The DSC is now the most popular and common assess- ages the loss function limit. Its value mainly depends on
ment indicator for assessing the segmentation results the gradient (negative) towards the model minima. The
and their base facts. This measures the overlap values training performance of the proposed improved ResNet
of two pairs, Q1 and Q2, via normalizing them well and existing CNN and FCN is described in Figure 6.
across the average of respective standard sizes. DSC is The proposed enhanced ResNet model shows a lower
presented in the equation error rate and higher accuracy in the training phase than
existing methods. The proposed improved ResNet model
TN
Specificity = (8) is validated using thirty percent of the training dataset in
TN + FP this experiment.

Testing results
Sensitivity and Specificity Figure 7 represents the performance validation of the
The following Eqs. 9 and 10 calculate sensitivity and proposed improved ResNet model with 50 epochs.
specificity as rule-based decision theory measures. Experimental outcomes prove that the training error rate
Where: TP-True Positive, FP-False Positive, TN-True decreases linearly, and the accuracy percentage increases
Negative, FN -False Negative for each epoch. The test dataset is implemented to the
TP proposed and existing model through the testing phase
Sensitivity = (9) to identify the brain Tumor cells in MRI images. The pro-
TP + FN
posed improved ResNet model is compared to specific
other existing methods in terms of performance metrics
TN (T, ET, WT) to analyze the performance of Tumor seg-
Specificity = (10)
TN + FP mentation. All performance measures have been taken
for each patient in the given dataset. The mean values of

Results
Training results Table 1 Training parameters of the proposed improved ResNet
In this research, the BraTS2020 dataset has been used model
collected from Kaggle [35]. This dataset mainly con-
Phase /steps Hyperparameter Parameters value
tains 369 brain Tumor patient MR images, where 125
are utilized for training and 169 MRI images for testing. Initialisation step Bias 0.1
The proposed improved ResNet model, existing CNN Weights Xavier
model, and FCN (model type U Net) are implemented ReLU (α) 0.333
using Python programming (Tensor flow) in the Ana- Drop out block LGG 0.111
conda environment. A complete experimental process HGG 0.555
is divided into two phases: training and testing. The Training step Number of Epochs for 50
first training phase is applied to train the model. LGG and HGG
In the first phase, the normalization process is used. Batch size 128
The dataset was corrected in the initial stage because Initial € value 0.004
the dataset had some inclination sub-field contortion Final € value 0.00004
for which the N4ITK technique has been taken. This Post Processing stage Batch Size 128
technique mainly converts all four MRI brain Tumor Tvol-HGG value 10,000
Tvol-HGG value 3,000
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 9 of 12

Fig. 6 Experimental outcomes for training accuracy of proposed improved ResNet and existing CNN and FCN

Fig. 7 Experimental outcomes for training Error Rate of proposed improved ResNet and existing CNN and FCN

these performance measures were then calculated for all

patients. Figure 8 shows the experimental results of the
proposed Improved ResNet Mode.

Discussions
Brain Tumor segmentation and detection is a widely
known area of research. Various Deep learning models
have been executed for all brain Tumor cases like core
Tumor region(CT), enhanced Tumor region(ET) and
whole Tumor region(WT).
The proposed Improved ResNet model is based on
Linked, which further performs identity mapping, and
one “s outcome is merged with the outcome of the con-
volution layer without using any model factors. It also
implies that a layer in the ResNet prototype tries to
understand the residual of interconnects.
In contrast, layers in CNNs and perhaps FCN (U-Net)
methods discover the actual performance. Consequently,
the gradients can move quickly back, leading to faster
computation than CNNs and FCN models. The quick
access links in the proposed Improved ResNet model
regulate the disappearing gradient issue. Fig. 8 Experimental Results of proposed Improved ResNet Mode
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 10 of 12

Table 2 Comparison of Existing and proposed improved ResNet Table 4 Comparison of Existing and proposed improved ResNet
model for Core Tumor Region (CT) model for Whole Tumor Region (WT)
Core Tumor Region (CT) Whole Tumor Region (WT)
Performance Existing CNN Existing FCN Proposed Performance Existing CNN Existing FCN Proposed
Measuring Model Model Improved Measuring Model Model Improved
Parameter ResNet Parameter ResNet

JC 0.6485 0.6225 0.658 JC 0.6695 0.6785 0.6308

DICE Score 0.9245 0.889 0.924 DICE Score 0.879 0.874 0.864
Sensitivity 0.7815 0.7256 0.7613 Sensitivity 0.7648 0.7465 0.7365
Specificity 0.831 0.814 0.835 Specificity 0.854 0.846 0.923
Accuracy 0.814 0.789 0.854 Accuracy 0.825 0.826 0.879

Tables 2, 3, and 4 compare proposed ResNet and Table 5 Experimental results of Existing and proposed improved
existing models (CNN and FCN) for JC, DICE Score, ResNet model for Enhanced Tumor Region (ET)
and Sensitivity, Specificity, and Accuracy param- Performance Existing CNN Existing FCN Proposed
eters for CT, ET and WT respectively on BraTS2020 Measuring Model Model Improved
datasets. Parameter ResNet
According to the assessment conducted for CT pro- MSE 28.647 33.9478 26.898
posed model, the output is 0.658, 0.924, 0.7613, 0.835, PSNR 30.789 29.898 21.457
and 0.854 of JC, DICE Score, Sensitivity, Specificity Computation Time 112 214 74
and Accuracy, respectively. Similarly, the ET proposed (in Minutes)
model is 0.6328, 0.945, 0.7989, 0.926, 0.913, and for
WT, it gives 0.6308, 0.864, 0.7365, 0.923, 0.879 values.
These results show improvement over CNN and
Conclusion & future work
FCN due to the four-phase process of the proposed
Deep Neural Networks (DNNs) are very useful for
model. The proposed Improved ResNet Model has
image segmentation. However, this technique encoun-
better outcomes for all three Tumor cases (ET, CT,
ters a disappearing gradient issue that emerges through-
and WT). This proves that the proposed Improved
out the training. To address this issue, the Improved
ResNet model performs well in pediatric segmenta-
ResNet is proposed in this research. A “connection link”
tion for a brain Tumor. Table 5 demonstrates that the
inside a current ResNet allows the gradient to propagate
proposed Improved ResNet model has the lowest com-
backwards to subsequent layers. These links provide all
putation time and the best PSNR and MSE. The pro-
the possible route details in a single place and provide
posed method has better results for MSE and PSNR
access in a single click reducing the accessing time. This
than existing CNN and FCN methods. Loewe, the
paper presents a pre-processing approach in the pro-
MSE value shows better performance. The proposed
posed method to eliminate many irrelevant data, result-
method has 26. 898% MSE and 21.457% PSNR are
ing in impressive outcomes.
more than 20%, far better than CNN and FCN.
The proposed Improved ResNet and existing CNN
and FCN models are implemented using tensor flow

Table 3 Comparison of Existing and proposed improved ResNet model for Enhanced Tumor Region (ET)
Enhanced Tumor Region (ET)
Performance Measuring Parameter Existing CNN Model Existing FCN Model Proposed
Improved
ResNet

JC 0.6515 0.6645 0.6328

DICE Score 0.941 0.895 0.945
Sensitivity 0.7989 0.74589 0.7989
Specificity 0.854 0.865 0.926
Accuracy 0.854 0.814 0.913
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 11 of 12

and tested on the BraTS2020 dataset. Experimen- Declarations

tal results demonstrate the strength of the proposed
Ethics approval and consent to participate
method in terms of better accuracy, less computation Not applicable.
time, MSE, PSNR, and better DSC and JC. The strength
of the proposed improved ResNet model is that users Consent for publication
Not applicable.
did not require the assistance of an expert to manually
find the Tumor pixel by pixel, which is a complex and Competing interests
time-consuming operation. This proposed model tack- The corresponding author here declares that there is no conflict of interest
from the other co-authors, including themselves.
les these issues by utilizing shortcut connection links
in ResNet.
The experimental outcomes achieve better perfor- Received: 16 December 2022 Accepted: 12 April 2023

mance and a remarkable result compared with conven-

tional techniques. In the binary classification problem,
accuracy and precision were examined, as was the Dice
References
coefficient score throughout the segmentation experi- 1. A Tiwari A, Srivastava S, Pant M. Brain Tumor segmentation and classifica-
ment. Future research can improve current outcomes tion from magnetic resonance images: Review of selected methods from
and leverage deeper architectures to improve the over- 2014 to 2019. Pattern Recognition Letters. 2020;131:244–60. https://​doi.​
org/​10.​1016/j.​patrec.​2019.​11.​020
all effectiveness of segmentation output. 2. Munir K, Frezza F, Rizzi A. Brain Tumor segmentation using 2D-UNET
convolutional neural network. Deep Learning for Cancer Diagnosis.
2021:239–48. https://​doi.​org/​10.​1007/​978-​981-​15-​6321-8_​14
Abbreviations 3. Aher P, Lilhore U. Survey of brain Tumor image quarrying techniques. Int J
MRI Magnetic resonance image Sci Eng Dev Res, ISSN. 2020:2455–631.
DNN Deep Neural Networks 4. Zhang D, Huang G, Zhang Q, Han J, Han J, Yu Y. Cross-modality deep
ResNet Residual Network feature learning for brain Tumor segmentation. Pattern Recogn.
FCN Fully Convolution Neural Network 2021;1(110). https://​doi.​org/​10.​1016/j.​patcog.​2020.​107562
VGG Visual Graphic group 5. Silva CA, Pinto A, Pereira S, Lopes A. Multi-stage deep layer aggregation
RL Residual learning value for brain Tumor segmentation. InBrainlesion: Glioma, Multiple Sclerosis,
CT Core Tumor Region Stroke, and Traumatic Brain Injuries: 6th International Workshop, BrainLes
MSE Mean Square Error 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 4, 2020,
JC Jaccard Coefficient Revised Selected Papers, Part II 6 2021 (pp. 179–188). Springer Interna-
MR Magnetic Resonance tional Publishing. https://​doi.​org/​10.​1007/​978-3-​030-​72087-2_​16
PET Positron emission tomography 6. Zhou T, Canu S, Vera P, Ruan S. Feature-enhanced generation and multi-
TP True Positive modality fusion based deep neural network for brain Tumor segmenta-
FP False Positive tion with missing MR modalities. Neurocomputing. 2021;27(466):102–12.
TN True Negative https://​doi.​org/​10.​1016/j.​neucom.​2021.​09.​032.
FN False Negative 7. Lin F, Wu Q, Liu J, Wang D, Kong X. Path aggregation U-Net model for
WT Whole Tumor Region brain Tumor segmentation. Multimedia Tools Appl. 2021;80:22951–64.
ET Enhanced Tumor Region https://​doi.​org/​10.​1007/​s11042-​020-​08795-9.
PSNR Peak Signal Noise Ratio 8. Das S, Swain MK, Nayak GK, Saxena S. Brain Tumor segmentation from
DSC Dice Similarity Coefficient 3D MRI slices using cascaded convolutional neural network. Advances
SGDOM Stochastic Gradient Descent optimization method in Electronics, Communication, and Computing: Select Proceedings of
RELM Regularized Extreme Learning Machine ETAEERE 2020 2021 (pp. 119–126). Springer Singapore. https://​doi.​org/​10.​
1007/​978-​981-​15-​8752-8_​12
Acknowledgements 9. Zhang Y, Lu Y, Chen W, Chang Y, Gu H, Yu B. MSMANet: a multi-scale mesh
We pay sincere thanks to all cited researchers. aggregation network for brain Tumor segmentation. Appl Soft Comput.
2021;1(110):107733. https://​doi.​org/​10.​1016/j.​asoc.​2021.​107733
Authors’ contributions 10. Munir K, Frezza F, Rizzi A. Deep learning for brain Tumor segmentation.
MA: writing and implementation of the proposed algorithm, results gathering, Deep Learning for Cancer Diagnosis. 2021:189–201. https://​doi.​org/​10.​
manuscript writing, analysis and interpretation of data. AKT: Supervision, 1007/​978-​981-​15-​6321-8_​11
formal analysis, validation, editing. MPS: formal analysis, critical manuscript 11. Vaibhavi P, Rupal K. Brain Tumor Segmentation Using K-means–FCM
revision, investigation, editing. AB: BraTS data set analysis, investigation, valida- Hybrid Technique. InAmbient Communications and Computer Systems:
tion, writing literature—review and editing. All authors read and approved the RACCCS 2017 2018 (pp. 341–352). Springer Singapore. https://​doi.​org/​10.​
final manuscript. 1007/​978-​981-​10-​7386-1_​30
12. Sharif MI, Li JP, Amin J, Sharif A. An improved framework for brain
Funding Tumor analysis using MRI based on YOLOv2 and convolutional neural
No External Funding has been received for this research from any International network. Complex Intell Syst. 2021;7:2023–36. https://​doi.​org/​10.​1007/​
or national body. s40747-​021-​00310-3.
13. Saueressig C, Berkley A, Munbodh R, Singh R. A joint graph and image
Availability of data and materials convolution network for automatic brain Tumor segmentation. In: Brain-
This work utilizes the online brain Tumor available dataset data from the Kag- lesion: Glioma, Multiple Sclerosis, Stroke, and Traumatic Brain Injuries: 7th
gle BraTS2020 competition. The following is the link: https://​www.​kaggle.​com/​ International Workshop, BrainLes 2021, Held in Conjunction with MICCAI
datas​ets/​awsaf​49/​brats​20-​datas​et-​train​ing-​valid​ation (accessed on 13 March 2021, Virtual Event, September 27, 2021, Revised Selected Papers, Part I.
2022). Cham: Springer International Publishing; 2022. p. 356–65. https://​doi.​org/​
10.​1007/​978-3-​031-​08999-2_​30.
Aggarwal et al. BMC Medical Informatics and Decision Making (2023) 23:78 Page 12 of 12

14. Zeineldin RA, Karar ME, Coburger J, Wirtz CR, Burgert O. DeepSeg: deep 33. Punn NS, Agarwal S. Multi-modality encoded fusion with 3D inception
neural network framework for automatic brain Tumor segmentation U-net and decoder model for brain Tumor segmentation. Multimedia
using magnetic resonance FLAIR images. Int J Computer-Assisted Radiol tools and applications. 2021;80(20):30305–20. https://​doi.​org/​10.​1007/​
Surg. 2020;15:909–20. https://​doi.​org/​10.​1007/​s11548-​020-​02186-z. s11042-​020-​09271-0.
15. Abd El Kader I, Xu G, Shuai Z, Saminu S, Javaid I, Salim Ahmad I. Differen- 34. Havaei M, Davy A, Warde-Farley D, Biard A, Courville A, Bengio Y, Pal C,
tial deep convolutional neural network model for brain Tumor classifica- Jodoin PM, Larochelle H. Brain Tumor segmentation with deep neural
tion. Brain Sci. 2021;11(3):352. https://​doi.​org/​10.​3390/​brain​sci11​030352. networks. Med Image Anal. 2017;1(35):18–31. https://​doi.​org/​10.​1016/j.​
16. Deng W, Shi Q, Luo K, Yang Y, Ning N. Brain Tumor segmentation based media.​2016.​05.​004.
on improved convolutional neural network in combination with non- 35. Online Kaggle Brain Tumor dataset. BraTS2020 Dataset (Training + Valida-
quantifiable local texture feature. J Med Syst. 2019;43:1–9. https://​doi.​org/​ tion). 2022. p. 13.
10.​1007/​s10916-​019-​1289-2. 36. Sharif MI, Li JP, Khan MA, Saleem MA. Active deep neural network features
17. Bodapati JD, Shaik NS, Naralasetti V, Mundukur NB. Joint training of selection for segmentation and recognition of brain Tumors using MRI
two-channel deep neural network for brain Tumor classification. SIViP. images. Pattern Recogn Lett. 2020;1(129):181–9. https://​doi.​org/​10.​1016/j.​
2021;15(4):753–60. https://​doi.​org/​10.​1007/​s11760-​020-​01793-2. patrec.​2019.​11.​019.
18. Zhou Z, He Z, Jia Y. AFPNet: A 3D fully convolutional neural network with 37. Singh K, Lilhore U, Agrawal N. Survey on different Tumor detection
atrous-convolution feature pyramid for brain Tumor segmentation via methods from MR images. Int J Sci Res Comput Sci Eng Inf Technol.
MRI images. Neurocomputing. 2020;18(402):235–44. https://​doi.​org/​10.​ 2017;5:589–94.
1016/j.​neucom.​2020.​03.​097. 38. Ghassemi N, Shoeibi A, Rouhani M. Deep neural network with generative
19. Jiang Y, Ye M, Huang D, Lu X. AIU-Net: An Efficient Deep Convolu- adversarial networks pre-training for brain Tumor classification based on
tional Neural Network for Brain Tumor Segmentation. Math Probl Eng. MR images. Biomed Signal Process Control. 2020;1(57):101678.https://​doi.​
2021;4(2021):1–8. https://​doi.​org/​10.​1155/​2021/​79157​06. org/​10.​1016/j.​bspc.​2019.​101678
20. Díaz-Pernas FJ, Martínez-Zarzuela M, Antón-Rodríguez M, González- 39. Saouli R, Akil M, Kachouri R. Fully automatic brain Tumor segmentation
Ortega D. A deep learning approach for brain Tumor classification and using end-to-end incremental deep neural networks in MRI images.
segmentation using a multi-scale convolutional neural network. Health- Comput Methods Programs Biomed. 2018;1(166):39–49. https://​doi.​org/​
care. 2021;9(2):153. https://​doi.​org/​10.​3390/​healt​hcare​90201​53. MDPI. 10.​1016/j.​cmpb.​2018.​09.​007.
21. Saleem H, Shahid AR, Raza B. Visual interpretability in 3D brain Tumor 40. Simaiya S, Lilhore UK, Prasad D, Verma DK. MRI brain Tumor detec-
segmentation network. Comput Biol Med. 2021;1(133):104410.https://​ tion & image segmentation by hybrid hierarchical K-means clustering
doi.​org/​10.​1016/j.​compb​iomed.​2021.​104410 with FCM-based machine learning model. Ann Roman Soc Cell Biol.
22. Gupta S, Gupta M. Deep learning for brain Tumor segmentation using 2021;28:88–94.
magnetic resonance images. In2021 IEEE conference on computational 41. Jia Q, Shu H. Bitr-unet: a cnn-transformer combined network for MRI brain
intelligence in bioinformatics and computational biology (CIBCB) 2021 Tumor segmentation. In: Brain lesion: Glioma, Multiple Sclerosis, Stroke
(pp. 1–6). IEEE. https://​doi.​org/​10.​1109/​CIBCB​49929.​2021.​95628​90 and Traumatic Brain Injuries: 7th International Workshop, Brain Les 2021,
23. Kamnitsas K, Ferrante E, Parisot S, Ledig C, Nori AV, Criminisi A, Rueckert Held in Conjunction with MICCAI 2021, Virtual Event, September 27, 2021,
D, Glocker B. DeepMedic for brain Tumor segmentation. InBrainlesion: Revised Selected Papers, Part II. Cham: Springer International Publishing;
Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries: Second 2022. p. 3–14. https://​doi.​org/​10.​1007/​978-3-​031-​09002-8_1.
International Workshop, BrainLes 2016, with the Challenges on BRATS,
ISLES and mTOP 2016, Held in Conjunction with MICCAI 2016, Athens,
Greece, October 17, 2016, Revised Selected Papers 2 2016 (pp. 138–149). Publisher’s Note
Springer International Publishing. https://​doi.​org/​10.​1007/​978-3-​319-​ Springer Nature remains neutral with regard to jurisdictional claims in pub-
55524-9_​14 lished maps and institutional affiliations.
24. Hao K, Lin S, Qiao J, Tu Y. A generalised pooling for brain Tumor seg-
mentation. IEEE Access. 2021;23(9):159283–90. https://​doi.​org/​10.​1109/​
ACCESS.​2021.​31300​35.
25. Iqbal S, Ghani MU, Saba T, Rehman A. Brain Tumor segmentation in multi-
spectral MRI using convolutional neural networks (CNN). Microsc Res
Tech. 2018;81(4):419–27. https://​doi.​org/​10.​1002/​jemt.​22994.
26. Isensee F, Jäger PF, Full PM, Vollmuth P, Maier-Hein KH. nnU-Net for brain
Tumor segmentation. InBrainlesion: Glioma, Multiple Sclerosis, Stroke and
Traumatic Brain Injuries: 6th International Workshop, BrainLes 2020, Held
in Conjunction with MICCAI 2020, Lima, Peru, October 4, 2020, Revised
Selected Papers, Part II 6 2021 (pp. 118–132). Springer International
Publishing. https://​doi.​org/​10.​1007/​978-3-​030-​72087-2_​11
27. Liu H, Li Q, Wang IC. A deep-learning model with learnable group convo-
lution and deep supervision for brain Tumor segmentation. Math Probl
Eng. 2021;10(2021):1–1. https://​doi.​org/​10.​1155/​2021/​66610​83.
28. Ramesh TR, Lilhore UK, Poongodi M, Simaiya S, Kaur A, Hamdi M. Predic-
tive analysis of heart diseases with machine learning approaches. Malays
Ready to submit your research ? Choose BMC and benefit from:
J Comput Sci. 2022;31:132–48. https://​doi.​org/​10.​22452/​mjcs.​sp202​2no1.​
10.
• fast, convenient online submission
29. Chen S, Ding C, Liu M. Dual-force convolutional neural networks for
accurate brain Tumor segmentation. Pattern Recogn. 2019;1(88):90–100. • thorough peer review by experienced researchers in your field
https://​doi.​org/​10.​1016/j.​patcog.​2018.​11.​009. • rapid publication on acceptance
30. Wadhwa A, Bhardwaj A. Verma VS A review on brain Tumor segmentation
• support for research data, including large and complex data types
of MRI images. Magn Reson Imaging. 2019;1(61):247–59. https://​doi.​org/​
10.​1016/j.​mri.​2019.​05.​043. • gold Open Access which fosters wider collaboration and increased citations
31. Lilhore U, Kumar S, Simaiya D, Prasad K. A Hybrid Tumor detection and • maximum visibility for your research: over 100M website views per year
classification based on machine learning. J Comput Theor Nanosci.
2020;17(6):2539–44. https://​doi.​org/​10.​1166/​jctn.​2020.​8927. At BMC, research is always in progress.
32. Wang Y, Peng J, Jia Z. Brain Tumor segmentation via c-dense convolu-
tional neural network. Progress in Artificial Intelligence. 2021;10:147–56. Learn more biomedcentral.com/submissions
https://​doi.​org/​10.​1007/​s13748-​021-​00232-8.