Transforming Brain Tumor Diagnosis with IVUM-Net: An Inclusive

Model for MRI-Based Detection and Classification

Tejaswi Murarry Setty, Bodagala Lakshmi Devi, K. Haripriya, S. Farhanabhanu,

M. Gurudhanush and G. Chandrasekar

Department of Electronics & Communication Engineering, Annamacharya Institute of Technology & Sciences, Rajampet,

Kadapa District, Andhra Pradesh, India

Keywords: Brain Tumor, MRI, Deep Learning, Artificial Intelligence, Medical Images, IVUM-Net, Convolution Neural

Networks (CNN), Transfer Learning.

Abstract: The accurate brain tumor diagnosis that occurs within proper time intervals ensures better patient care and

treatment success. The research works to create IVUM-Net which represents an innovative AI model to

improve brain cancer detection along with classification using MRI information. Advanced digital image

processing methods with Convolutional Neural Networks help the proposed model conduct automated tumor

detection practice. IVUM-Net leverages the capabilities of Inception V3 for feature extraction, U-Net for

accurate segmentation, and Multi-Class Support Vector Machine (MCSVM) for robust classification. Data

augmentation together with transfer learning methods will optimize performance levels and preprocessing

methods will optimize picture quality for the model. The method aims to eliminate human mistakes in

addition to reducing the need for visual assessment. Class activation mapping (CAM) serves as an

interpretability tool by visualizing how the model decides between classes. The research aims at verifying

IVUM-Net as an effective medical instrument for early brain tumor diagnosis and classification procedures

to enhance treatment approaches.

1 INTRODUCTION

Neurological patients need both early detection of

brain tumors and precise identification to get better

therapeutic results. MRI takes the lead as a common

non-invasive method because of its detailed imaging

ability in brain tumor detection. The human

interpretation of MRI scans requires extensive time

commitment and produces risks of misdiagnosis and

treatment delays because of human error. The

preference for diagnosis of brain tumors forms around

MRI which remains the most commonly used

technique. Deep learning models from artificial

intelligence have demonstrated substantial potential

to detect tumors while performing classification due

to the rising demand for better diagnosis methods.

The research work presents IVUM-Net as an

advanced AI-based system which enhances

automated brain tumor detection through MRI

analysis. IVUM-Net unifies MCSVM with its

dependable classification capabilities together with

CNNs to accomplish feature extraction and U-Net

capabilities that enable precise segmentation.

Effective feature extraction, and U-Net for precise

segmentation. Data augmentation that incorporates

transfer learning strategies achieves better

performance and generalization along with

preprocessing methods in this model.This model

works toward decreasing manual interpretation needs

while reducing human mistakes and providing faster

and clearer brain tumor detection that supports

improved treatment planning for patient care.

The diagnosis of multiple brain tumor types

depends on Support Vector Machines (SVMs) which

operate alongside deep learning systems. Using

Multi-Class Support Vector Machines (MCSVM)

provides medical developers with a dependable

method for classifying diverse tumor types to achieve

effective differentiation of multiple tumor types.

SVMs demonstrate robust generalization power

which applies favorably to medical image

categorization needs thus enabling their incorporation

into artificial intelligence brain tumor

detection platforms.

Setty, T. M., Devi, B. L., Haripriya, K., Farhanabhanu, S., Gurudhanush, M. and Chandrasekar, G.

Transforming Brain Tumor Diagnosis with IVUM-Net: An Inclusive Model for MRI-Based Detection and Classiﬁcation.

DOI: 10.5220/0013919000004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 4, pages

683-689

ISBN: 978-989-758-777-1

683

Doctors can use Class Activation Mapping

(CAM) technology to see where in the input image

the model places its main emphasis for prediction

purposes thus raising their trust in AI diagnostic

systems. Various investigations demonstrate how AI-

based models show excellent efficiency together with

high accuracy in brain tumor diagnosis. The

successful operation of high-performing models

alongside their interpretability presents an ongoing

challenge because current systems have difficulty

working across different dataset and imaging

conditions. The research presents IVUM-Net as a

hybrid model which combines Convolutional Neural

Networks (CNNs) along with U- Net and MCSVM to

solve precise brain tumor detection and classification

needs with preprocessing techniques and data

augmentation and transfer learning components.

2 LITERATURE REVIEW

Researchers have shown significant interest in recent

times regarding the implementation of artificial

intelligence in medical imaging to detect brain

tumors. Medical image analysis automation occurs

from the implementation of machine learning

methodologies with deep learning methods using

Convolutional Neural Networks (CNNs). The

exceptional capability of CNNs for hierarchy

extraction from image data makes them valuable tools

in detecting tumors and performing their

classification. Many research studies have proven

how CNN-based systems recognize brain tumors

from normal tissue structures in MRI image data.

The U-Net model proves superior to other

segmentation models because it executes pixel-wise

segmentation with the critical requirement to

accurately define tumors. The segmentation process

of U-Net benefits from both encoder-decoder

structures alongside skip connections which maintain

spatial information. The medical imaging application

of U-Net has led to numerous brain tumor

segmentation procedures and researchers utilize CNN

integration to boost brain MRI tumor segmentation

abilities.

K. P. Bedi and J. S. Jadon from 2024 performed

their research on deep learning methods to identify

brain tumors through MRI image processing

applications. The top model achieved 94.7% accuracy

together with 93.9% specificity in its performance.

System results were affected by how design

components and dataset structures interacted

according to this research finding.

In 2023 R. Mishra developed a brain tumor

detection system based on the Robust Active Shape

Model Algorithm operating within a deep learning

architecture. The detection method showed precision

of 93.5% and 92.8% specific detection performance.

The detection system showed capability in processing

tumors with multiple forms along with various

shapes.

V. Kushwaha and P. Maidamwar conducted 2022

research to evaluate the SVR and CNN-based

machine learning techniques for brain tumor

identification using experimental experimental

approaches. The methodology reported 92.4%

accuracy together with 91.2% specificity as its major

performance metrics. The selection of suitable

algorithms leads to maximum result performance

based on this research analysis.

Brain tumor MRI image classification received

deep transfer learning treatment in 2021 according to

the research from O.P. Özlem and C. Güngen.

Medical imaging received confirmation of its

effectiveness because the method achieved 93.7%

accuracy while observing 92.5% specificity. The

applied approach reduced the need for large training

dataset quantities.

In 2020 H. A. Khalil together with coauthors

presented a 3D-MRI brain tumor detection system

which combined modified level set segmentation

with the dragonfly algorithm. The model evaluation

showed 92.8% accuracy and 91.6% specificity as key

results. Better clarity of segmentation coupled with

reduced computational complexity arose from the

combination of these two components.

In 2021 researcher Ö. P. Özlem and C. Güngen

applied deep transfer learning for brain tumor

classification on MRI images through pre-trained

network optimization. The study established 93.7%

accuracy with 92.5% specificity thereby proving

transfer-learning is an effective solution for medical

imaging tasks. The approach needed minimal

information about training data for practitioners in

healthcare to successfully carry out their work.

Research done by H.A. Khalil and colleagues in

2020 resulted in a 3D-MRI brain tumor detection

system through integration of the dragonfly algorithm

with modified level set segmentation. The developed

prototype demonstrated 92.8% accuracy combined

with 91.6% specificity. The combination of these

methods produced superior segmentation results

through an operation system that needed fewer

processing capabilities.

Z. Huang together with colleagues conducted

brain tumor classification research using a CNN-

based model which became more efficient through

activation function modification. The accuracy rate of

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

684

the study reached 94.1% without losing 93.3%

specificity. The model received activations function

modifications to boost its ability to match specific

features and enhance its classification outcome

performance.

P. M. Krishnammal and S. S. Raja established a

CNN-based detection system for brain abnormalities

in MRI images through their research work during

2019. The system delivered detection results with

94.8% accuracy and 95.2% specific outcomes. The

detection method proved to have outstanding

reliability in identifying abnormal conditions.

Zhou et al created UNet++ as a medical image

segmentation architecture that improved output

resolution through redesigned skip connections

during 2018. The developed model demonstrated

outstanding performance with 96.5% accuracy

together with 95.4% specificity and its optimal results

were achieved during complex medical structure

segmentation. The network architecture proposed

information that fixed various problems existing in

classic U-Net systems.

3 METHODOLOGY

3.1 Existing method

3.1.1 Convolutional Neural Networks

(CNNs)

It demonstrates effectiveness in delivering top-level

performance while handling various kinds of

applications. The method operates with a CNN

foundation that develops refined features from

different data types through their spatial organization

to generate accurate results.

A basic CNN consists of numerous convolutional

and pooling layers which come before the fully

connected layers. The training process utilizes

significant collections of labeled data samples that

optimize a loss metric which measures forecast versus

observed value differences.

The CNN-based method detects tumor through

an analysis of image color and texture.

The process of color analysis extracts visual color

attributes from pictures for the recognition of skin

discolorations.

The process of texture pattern analysis assesses

skin appearances for the purpose of separating

healthy and affected tissue sections.

For skin area separation the CNN framework uses

segmentation procedures. This process includes:

The technique sorts images according to

respective interest zones for diagnosing areas with

tumors.

Masks are created by this method to enhance

tumor areas and simultaneously decrease the

visibility of normal skin tissue. Figure 1 shows the

IVUM-Net Architecture Flowchart for Brain Tumor

Detection.

3.2 Proposed Method

Figure 1: IVUM-Net Architecture Flowchart for Brain

Tumor Detection.

3.2.1 Input Block (Mri Scans)

• MRI Data Input: This part takes in MRI images

of the brain, which show areas that may have

tumors. These images are the main input for the

model to work with.

3.2.2 Preprocessing Block

• Image Quality Enhancement: This step

enhances MRI visual information through the

elimination of distortions while it optimizes the

pictures by adjusting brightness alongside

contrast parameters. Due to improved image

quality the identification features of tumors

become more visible to the model.

• Data Augmentation: This produces

supplementary MRI image examples through

Transforming Brain Tumor Diagnosis with IVUM-Net: An Inclusive Model for MRI-Based Detection and Classiﬁcation

685

flipping and rotation along with minor

modifications to the pictures. The model learns

more efficiently to detect tumors across multiple

image varieties through this process.

• Transfer Learning Preparation: This step

researchers enhance the MRI images so they can

acquire knowledge from existing models that

analyze particular features while speeding up

training and improving accuracy with fewer

images in collection.

3.2.3 Segmentation Block (U-Net)

• Encoder: Encodes function which processes

image features works progressively to decrease

dimensions and capture tumor location context at

a high level.

• Decoder: It utilizes image features for a pixel-

by-pixel mapping which leads to accurate tumor

segmentation.

• Skip Connections: The network transfers

encoder-coded information to the decoder for

more precise segmentation results.

3.2.4 Feature Extraction Block (Inception

V3)

• Convolutional Layers: It enable the model to

detect characteristic pattern arrangements in

MRI imaging data which may point out tumor

characteristics such as irregular shapes together

with unusual textures.

• Pooling Layers: These layers combine with

others to downscale image information while

selecting essential characteristics and discarding

superfluous information. The model obtains

simpler image processing because of this

technique.

3.2.5 Interpretability Block (Class

Activation Mapping - Cam)

• Class Activation Mapping: This part shows

which areas in the MRI image the model found

most important for making its decision. It

highlights these areas to help doctors understand

why the model thinks a tumor is a certain type.

3.2.6 Classification Block (Multi-Class Svm)

• Support Vector Machine (SVM) Layers the

Support Vector Machine (SVM) Layers serve as

the classifier which divides the segmented tumor

between different types. The classification block

evaluates tumors to identify their A, B

or C categories.

• Multi-Class Handling: The model can handle

multiple types of tumors, so it doesn’t just look

for one kind but can identify several types based

on what it has learned.3.2.

3.2.7 Output Block

• Diagnosis Result: The model provides its final

diagnosis, specifying the type of tumor detected.

• Segmentation Map: It displays tumor position

specifically in MRI images so physicians can

determine its precise area.

Table 1 shows the

Features Extracted.

• Interpretability Report: It provides visual

representations of areas in the MRI image that

the model used for making its diagnosis thereby

enhancing diagnostic transparency.

Table 2

shows the Comparision of Existing and Proposed

Algorithms.

Table 1: Features Extracted.

Samples Contrast Accuracy Energy Execution

Time

Glioma 0.4785 99.39% 0.358 1.30s

Metastasis 0.4469 95.73% 0.258 1.64s

Astrocytoma 0.383 96.38% 0.262 1.56s

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

686

Table 2: Comparision Of Existing and Proposed Algorithms.

S.No Criteria Existing Method: CNN Proposed Method: IVUM-Net

1 Advantages

1. Learns hierarchical

features from input data.

2. Suitable for image

analysis tasks.

1. Integrates Inception V3, U-Net, and MCSVM for

enhanced accuracy and performance.

2. Uses preprocessing, data augmentation, and

transfer learning for better robustness and image

quality.

3. Incorporates class activation mapping (CAM) for

roved inter

retabilit

of decision-makin

2 Disadvantages

1. Requires a large

labeled dataset for

training.

2. Susceptible to

overfitting with deep

networks.

3. May lack segmentation

precision and

trans

arenc

1. Computationally intensive due to the integration

of multiple techniques.

2. Complexity increases due to model design

combining CNN, U-Net, and MCSVM.

3. Performance depends on MRI scan quality and

preprocessing steps.

Expected

Performance

1. Efficient in feature

extraction.

2. Strong results in image

classification tasks.

1. Achieves better segmentation (U-Net) and

classification (MCSVM) for brain tumor detection.

2. Provides robust and accurate multi-class

classification with precise segmentation.

3. Reduces human error through automation and

uses CAM for decision transparency.

4 RESULT ANALYSIS

Figure 2: Input Image.

This figure 2 The image displays original MRI brain

data that maintains entire information along with both

valuable content and superfluous areas

and system noise.

Figure 3: Pre-Processed Image using IVUM-Net.

Figure 4: Segmented Image Using IVU-Net.

Transforming Brain Tumor Diagnosis with IVUM-Net: An Inclusive Model for MRI-Based Detection and Classiﬁcation

687

Figure 5: Detected Results.

Figure 6: Analysis Image.

The Analysis image shows a confusion matrix that

evaluates how well the system classifies different

types of brain tumors. The diagonal values represent

correct predictions: 248 Gliomas, 246 Meningiomas,

244 Metastases, and 244 Astrocytomas were

identified accurately. Misclassifications are minimal,

such as 2 Gliomas labeled as Meningiomas and 4

Meningiomas labeled as Metastases. The system

displays notable success through its accurate

performance although errors do exist. The system

maintains its high accuracy level which was

previously observed. Figure 3 shows the Pre-

Processed Image using IVUM-Net. Figure 4 shows

the Segmented Image Using IVUM-Net. Figure 6

shows the Analysis image. Table 3 shows the

Comparision Metrics.

5 PERFORMANCE

COMPARISON

Table 3: Comparision Metrics.

Method [author name] Accuracy Specificity

Classifying Brain

Tumors using CNN

[ Badža MM]

92.9% 91.7%

Comparison of Deep

Learning Methods

[K. P. Bedi and J. S.

Jadon,]

94.7% 93.9%

Robust Active Shape

Model Algorithm

93.5% 92.8%

Empirical Analysis of

ML Techniques

92.4% 91.2%

Deep Transfe

Learning 93.7% 92.5%

3D-MRI with Level Set

+ Dragonfly Algorithm

[

H. A. Khalil]

92.8% 91.6%

CNN with Modified

Activation Function

[

L. Chen, et al.,]

94.1% 93.3%

CNN-Based MRI Image

Classification

[

P. M. Krishnammal

S. S. Ra

a,]

94.8% 95.2%

IVUM-NET(proposed

method)

95.34% 96.0%

REFERENCES

Abdusalomov Ab, Mukhiddinov M, Whangbo Tk. Brain

Tumor Detection Based Magnetic Resonance Imaging

and Deep Learning Methods for Brain Tumor Detection.

Cancers (Basel). 2023 Aug 18;15(16):4172. Doi:

10.3390/Cancers 15164172. Pmid: 37627200; Pmcid:

Pmc10453020.

Badža MM, Barjaktarović MČ. Classification of brain

tumors from MRI images using a convolutional neural

network.Appl Sci. 2020;10(6):1999

Namya Musthafa, Mohammad Mehedy Masud and Qurban

Memon. 2024. Advancing Early-Stage Brain Tumor

Detection with Segmentation by Modified_UNet.

Health Informatics (ICMHI 2024), May 17-19, 2024,

Yokohama, Japan.ACM, New York, NY, USA,Pages.

https://doi.org/10.1145/3673971.3674001

Saeedi, S., Rezayi, S., Keshavarz, H. et al. MRI-based brain

tumor detection with selected machine learning

algorithms and convolutional deep learning

approaches.BMC Med Inform Decis Mak 23, 16 (2023).

https://doi.org/10.1186/s12911-023-02114-6

K. P. Bedi and J. S. Jadon, "Comparison of Deep Learning

Methods for Detecting Brain Tumor," 2024 14th

International Conference on Cloud Computing, Data

Science & Engineering (Confluence), Noida, India,

2024,pp.897901,doi:10.1109/Confluence60223.2024.1

0463287.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

688

H. A. Khalil, S. Darwish, Y. M. Ibrahim, and O. F. Hassan,

"3D-MRI brain tumor detection model using modified

version of level set segmentation based on dragonfly

algorithm," Symmetry, vol. 12, no. 1256, 2020.

Z. Huang, X. Du, L. Chen, et al., "Convolutional Neural

Network Complex Networks for Brain Tumor Image

Classification with a Modified Activation Function,"

IEEE Access, vol. 8, pp. 89281–89290, 2020.

P. M. Krishnammal and S. S. Raja, "Convolutional neural

network-based image classification and detection of

abnormalities in M.R.I. brain images," in Proc. 2019

Int. Conf. Communication and Signal Processing

(ICCSP), Melmaruvathur, Tamil Nadu, India, 4–6 Apr.

2019.

Transforming Brain Tumor Diagnosis with IVUM-Net: An Inclusive Model for MRI-Based Detection and Classiﬁcation

689