DATA SCIENCE

A SUMMER INTERN REPORT

Submitted by

AMAN BHARDWAJ
08614802821

in partial fulfilment of Summer Internship for the award of the degree

of

BACHELOR OF TECHNOLOGY
IN

ELECTRONICS AND COMMUNICATION

Maharaja Agrasen Institute of Technology

1
BRAIN TUMOR DETECTION USING TRANSFER
LEARNING
A SUMMER TRAINING REPORT

Submitted by

AMAN BHARDWAJ

Enrolment Number: 08614802821

Electronics and Communication Engineering

Under the supervision of

ABHIJEET BHATACHARYA
TEAM LEAD

CODING BLOCKS

DELHI

MAHARAJA AGRASEN INSTITUTE OF TECHNOLOGY

ROHINI, NEW DELHI

Maharaja Agrasen Institute of Technology

2
 To Whom It May Concern
I, AMAN BHARDWAJ, Enrolment No. 08614802821, a student of Bachelors of Technology

(ECE), a class of 2021-25, Maharaja Agrasen Institute of Technology, Delhi hereby declare

that the Summer Training project report entitled “BRAIN TUMOR DETECTION USING

TRANSFER LEARNING” is an original work and the same has not been submitted to any

other Institute for the award of any other degree.

Date: 24-09-2023

Place: NEW DELHI

AMAN BHARDWAJ
Enrolment No: 08614802821
Electronics and Communication Engineering
5E123

ACKNOWLEDGEMENT

3
First and foremost, I wish to express my profound gratitude to VARUN KOHLI, CODING
BLOCKS, DELHI for giving me the opportunity to carry out my project. I find great pleasure to
express my unfeigned thanks to him, for his invaluable guidance, support and useful suggestions
at every stage of this project work.

No words can express my deep sense of gratitude to Mr. ABHIJEET BHATTACHARYA,

CODING BLOCKS, without whom this project would not have turned up this way. My heartfelt
thanks to him for his immense help and support, useful discussions and valuable
recommendations throughout the course of my project work.

I wish to thank my respected faculty and my lab mates for their support.

Last but not the least I thank the almighty for enlightening me with his blessings.

AMAN BHARDWAJ
Enrolment Number:08614802821
5E456/E4

4
 ABOUT THE COMPANY (CODING BLOCKS)

Coding Blocks was founded in 2014 with a mission to create skilled Software Engineers for our
country and the world. We are here to bridge the gap between the quality of skills demanded by
industry and the quality of skills imparted by conventional institutes. At Coding Blocks, we
strive to increase student interest by providing hands on practical training on every concept
taught in the classroom.

Coding Blocks has been built with a vision of creating Software Developers and Entrepreneurs
who can shape the future of our country. India has around 5 lac engineers graduating from
colleges every year, majority of which aren't employable. On the other hand most of the Software
Companies aren't able to fill their job vacancies because they don't get good candidates.

The problem lies in the fact that the content and methodology of teaching hasn't kept the pace
with the fast changing IT sector. Another problem is that no IT professionals want to
mentor/teach these students because teaching is far less lucrative financially compared to
working for IT firms.

At Coding Blocks we want to bridge this gap by enabling students with the right skills which
will help them by not just securing the best jobs but also enabling them to be productive from
day 1 of their corporate lives.

5
 ABSTRACT

This summer training report provides a comprehensive overview of the journey undertaken in the realm
of DATA SCIENCE. The training program encompassed a wide array of essential topics, equipping
participants with a solid foundation in machine learning, programming, data manipulation, visualization,
and statistical analysis.

The training began with an "Introduction to Machine Learning," delving into the core concepts and
principles that underpin this transformative field. Participants were then introduced to the versatile
Python programming language, and its libraries such as NumPy, Pandas, Matplotlib, Seaborn, and Plotly,
providing essential tools for data manipulation and visualization.

A firm grasp of statistics formed the basis for rigorous analysis, enabling trainees to make datadriven
decisions. The pivotal Scikit-Learn and tensorflow library was explored in detail, offering a wide range of
machine learning and deep learning models and tools for predictive and regressive analytics.

The highlight of the training was the implementation of machine learning algorithms, where participants
were guided through the process of building, training and evaluating models. This hands-on experience
culminated in a real-world project: "BRAIN TUMOR DETECTION USING TRANSFER LEARNING." Leveraging
their newfound knowledge, trainees successfully applied machine learning techniques to differentiate
between underwater objects, demonstrating the practical application of the skills acquired.

In conclusion, this summer training report encapsulates the holistic learning experience, bridging theory
and practice in the field of deep Learning. It serves as a testament to the participants' dedication and
their newfound proficiency in Python programming, data analysis, statistics, and machine learning, with
a real-world project showcasing their ability to address complex challenges through the power of data
and algorithms.

6
CERTIFICATE

7
 TABLE OF CONTENTS

S.NO. CHAPTER PAGE NO.

1) Introduction to artificial 9
intelligence

2) Introduction to machine 10-11

learning

3) Introduction to deep 11-13

learning

4) Introduction to transfer 16-18

learning

5) Convolutional neural 19-21

networks

6) Image processing 22-28

8
 CHAPTER-1
INTRODUCTION

1.1 INTRODUCTION TO ARTIFICIAL INTELLIGENCE (AI)

Artificial Intelligence (AI) is a rapidly evolving field of technology and computer science that
focuses on creating systems capable of performing tasks that typically require human
intelligence. These tasks include understanding natural language, recognizing patterns, solving
problems, learning from experiences, and making informed decisions.
AI seeks to simulate human-like intelligence in machines by enabling them to process vast
amounts of data, recognize patterns, and adapt to new information or stimuli. It encompasses a
broad range of techniques, approaches, and methodologies, with the ultimate goal of enabling
machines to imitate human cognition and problem-solving abilities.

Key components of AI include:

1. Machine Learning (ML): ML is a subset of AI that involves training machines to learn from
data and improve their performance on specific tasks without being explicitly programmed.

2. Deep Learning (DL): DL is a specialized form of ML that uses neural networks with multiple
layers to analyze data and extract intricate patterns, often achieving superior accuracy in various
applications.

3. Natural Language Processing (NLP): NLP focuses on enabling machines to understand,

interpret, and generate human language, bridging the gap between computers and human
communication.

9
4. Computer Vision (CV): CV involves enabling machines to interpret and understand visual
information from images or videos, akin to how humans perceive the world.

5. Robotics: AI in robotics involves creating intelligent machines capable of interacting with the
environment and making decisions based on sensory inputs.

INTRODUCTION TO ML

1.2 Introduction to Machine Learning (ML)

Machine Learning (ML) is a subset of artificial intelligence (AI) that focuses on enabling
computer systems to learn and improve from experience without being explicitly programmed
for each task. It involves the development of algorithms and models that allow machines to
automatically recognize patterns, make predictions, and optimize their performance based on
data.

Key concepts and components of ML include:

1. Data: Data is the foundation of ML. Algorithms learn patterns and make predictions based on
the data they are provided. Quality and quantity of data significantly influence the ML model's
effectiveness.

2. Training: ML models undergo a training phase where they learn patterns and relationships
within the provided data. During training, the model adjusts its parameters to minimize errors
and improve its accuracy in making predictions.

3. Features: Features are specific characteristics or attributes extracted from the data that help
the model understand and differentiate patterns. Choosing relevant features is crucial for accurate
predictions.

4. Algorithms: ML algorithms are mathematical and statistical techniques used to train models
and make predictions. These algorithms can be categorized into supervised learning,
unsupervised learning, reinforcement learning, and more, depending on the learning approach.

10
5. Model Evaluation and Testing: After training, models need to be evaluated on separate data
(testing data) to assess their performance and generalization abilities. This helps identify if the
model can make accurate predictions on unseen data.

ML finds applications in a wide range of domains, including but not limited to:
- Predictive Analytics: Predicting future outcomes based on historical data.
- Natural Language Processing (NLP): Understanding and processing human language.
- Computer Vision: Interpreting and analyzing visual information from images or videos.
- Recommendation Systems: Suggesting products or content based on user preferences.
- Healthcare: Diagnosing diseases and predicting patient outcomes.

ML has revolutionized industries, providing powerful tools to extract insights, automate

processes, and enhance decision-making. Its potential to uncover valuable patterns in data and its
continuous advancement fuel ongoing research, innovation, and integration into diverse
applications, shaping the future of technology and problem-solving.

INTRODUCTION TO DEEP LEARNING

1.3 DEEP LEARNING

Deep Learning is a subfield of artificial intelligence (AI) and machine learning (ML) that
involves training and using neural networks with multiple layers (hence "deep") to analyze and
extract intricate patterns from data. The objective of deep learning is to imitate the way the
human brain processes and understands information, enabling computers to learn, generalize, and
make predictions or decisions.
Recurrent neural networks (RNN) are very important here because they make it possible to
model time sequences instead of just considering input and output frames independently. This is
especially important for noise suppression because we need time to get a good estimate of the
noise.
Deep learning is the new version of an old idea: artificial neural networks. Although those have
been around since the 60s, what's new in recent years is that:

11
 1. We now know how to make them deeper than two hidden layers
2. We know how to make recurrent networks remember patterns long in the past
3. We have the computational resources to actually train them

Key Components of Deep Learning:

1. Neural Networks: These are the foundational building blocks of deep learning. They
consist of interconnected nodes (neurons) organized in layers. Each neuron receives
input, processes it using weights and activation functions, and passes the output to the
next layer.
2. Layers: Neural networks are organized into multiple layers, including an input layer,
hidden layers, and an output layer. Deep learning models typically have many hidden
layers that allow for the extraction of complex features.
3. Activation Functions: Activation functions introduce non-linearity into the model,
enabling it to learn complex patterns. Common activation functions include ReLU
(Rectified Linear Unit), Sigmoid, and Tanh.
4. Backpropagation: An optimization algorithm that adjusts the weights and biases of the
neural network during the training process. It calculates the gradient of the loss function
with respect to the weights using the chain rule and updates the parameters accordingly.

Training Deep Learning Models:

1. Data Preparation: Relevant data is collected, preprocessed, and split into training,
validation, and testing sets.
2. Model Architecture: Design the neural network architecture, including the number and
structure of layers, activation functions, and loss functions.
3. Training: The model is fed with training data, and it adjusts its weights and biases
iteratively through forward and backward passes using backpropagation to minimize the
loss function.
4. Validation and Tuning: The model's performance is evaluated on the validation set, and
adjustments are made to prevent overfitting (excessive focus on training data) or
underfitting (insufficiently learning from data).
5. Testing: Finally, the model is evaluated on a separate testing set to assess its
generalization and predictive capabilities.

12
Applications of Deep Learning:
 Computer Vision: Image and video recognition, object detection, segmentation, and
image generation.
 Natural Language Processing (NLP): Language translation, sentiment analysis,
chatbots, and text summarization.
 Speech and Audio Processing: Speech recognition, speaker identification, and audio
generation.
 Autonomous Vehicles: Self-driving cars use deep learning for perception and decision-
making.
 Healthcare: Disease diagnosis, drug discovery, and personalized medicine.

Advantages of Deep Learning:

 Feature Learning: Deep learning algorithms can learn hierarchical representations of
data, automatically extracting features at multiple levels.
 Scalability: Deep learning models can scale with the size of data, making them suitable
for big data analytics.
 State-of-the-Art Performance: Deep learning models often achieve or surpass human-
level performance in various tasks.

Deep learning has revolutionized AI, enabling breakthroughs in numerous domains. Continued
research and innovation in deep learning promise to unlock even more advanced capabilities and
applications.

INTRODUCTION TO TRANSFER LEARNING

1.4 TRANSFER LEARNING
The reuse of a pre-trained model on a new problem is known as transfer learning in machine
learning. A machine uses the knowledge learned from a prior assignment to increase prediction
about a new task in transfer learning. You could, for example, use the information gained during
training to distinguish beverages when training a classifier to predict whether an image contains
cuisine.

13
The knowledge of an already trained machine learning model is transferred to a different but
closely linked problem throughout transfer learning. For example, if you trained a simple
classifier to predict whether an image contains a backpack, you could use the model’s
training knowledge to identify other objects such as sunglasses.

How Transfer Learning Works?

In computer vision, neural networks typically aim to detect edges in the first layer,
forms in the middle layer, and task-specific features in the latter layers.

The early and central layers are employed in transfer learning, and the latter layers are
only retrained. It makes use of the labelled data from the task it was trained on.

Transfer learning offers a number of advantages, the most important of which are
reduced training time, improved neural network performance (in most circumstances),
and the absence of a large amount of data.

To train a neural model from scratch, a lot of data is typically needed, but access to
that data isn’t always possible – this is when transfer learning comes in handy.

Transfer learning is a powerful technique used in Deep Learning. By harnessing the

ability to reuse existing models and their knowledge of new problems, transfer
learning has opened doors to training deep neural networks even with limited data.
This breakthrough is especially significant in data science, where practical scenarios
often need more labeled data. In this article, we delve into the depths of transfer

14
learning, unraveling its concepts and exploring its applications in empowering data
scientists to tackle complex challenges with newfound efficiency and effectiveness.

Models That Have Been Pre-Trained

There are a number of popular pre-trained machine learning models available. The
Inception-v3 model, which was developed for the ImageNet “Large Visual
Recognition Challenge,” is one of them.” Participants in this challenge had to
categorize pictures into 1,000 subcategories such as “zebra,” “Dalmatian,” and
“dishwasher.”

End Note

In conclusion, understanding transfer learning is crucial for data scientists venturing

into deep learning. It equips them to leverage pre-trained models and extract valuable
knowledge from existing data, enabling them to solve complex problems with limited
resources.

15
 CHAPTER-2

TensorFlow: Empowering the World of Deep Learning

2.1 Introduction:

TensorFlow, an open-source machine learning library developed by the Google Brain team, has
emerged as a powerful tool in the realm of deep learning. Released in 2015, TensorFlow has
quickly become a cornerstone in the development and deployment of artificial intelligence (AI)
applications, providing a flexible and efficient platform for building and training various
machine learning models.

2.2 Core Concepts:

1. Tensors:
At the heart of TensorFlow is the concept of tensors, which are multi-dimensional arrays
representing the data manipulated by the computational graphs. Tensors flow through the graph,
capturing the transformations and operations applied to the data. This symbolic representation
facilitates efficient parallel computation on both CPUs and GPUs.

2. Computational Graphs:
TensorFlow utilizes a dataflow graph paradigm to represent computations. Nodes in the graph
represent operations, and edges represent the flow of tensors between these operations. This
graph-based approach allows for efficient optimization, distribution, and deployment of machine
learning models.

Key Features:
1. Flexibility and Scalability:
TensorFlow's flexibility is one of its standout features. It supports a wide range of tasks, from
traditional machine learning to complex deep learning models. Its modular architecture allows

16
researchers and developers to experiment with various model architectures, making it suitable for
both prototyping and production deployment. Additionally, TensorFlow's scalability enables the
training of large models on distributed computing systems.

2. High-level APIs:
TensorFlow offers high-level APIs that simplify the process of building and training models.
TensorFlow Keras, a high-level neural networks API, allows for easy construction of neural
networks with just a few lines of code. This abstraction makes it accessible for both beginners
and experienced practitioners, promoting rapid development and experimentation.

3. TensorFlow Lite and TensorFlow.js:

TensorFlow extends its reach to mobile and web applications through TensorFlow Lite and
TensorFlow.js, respectively. TensorFlow Lite enables the deployment of machine learning
models on mobile and embedded devices, catering to the growing demand for on-device
inference. TensorFlow.js, on the other hand, brings machine learning capabilities to web
browsers, fostering the development of interactive and intelligent web applications.

2.3 Use Cases:

1. Image and Speech Recognition:
TensorFlow has been instrumental in the development of state-of-the-art image and speech
recognition models. Applications ranging from facial recognition systems to voice assistants
leverage TensorFlow's capabilities to achieve high accuracy and efficiency.

2. Natural Language Processing (NLP):

TensorFlow powers numerous natural language processing applications, including language
translation, sentiment analysis, and chatbot development. Models like BERT and GPT, built on
TensorFlow, have set new benchmarks in understanding and generating human-like text.

3. Healthcare and Biotechnology:

In healthcare, TensorFlow has found applications in medical image analysis, disease prediction,
and drug discovery. The ability to process vast amounts of medical data and extract meaningful
insights has the potential to revolutionize the diagnosis and treatment of various medical
conditions.

17
4. Autonomous Systems:
TensorFlow plays a pivotal role in the development of machine learning models for autonomous
systems, including self-driving cars and drones. Its ability to process real-time data and make
split-second decisions is crucial for the safety and efficiency of these systems.

Future Directions:
As the field of machine learning continues to evolve, TensorFlow remains at the forefront of
innovation. Ongoing developments include enhancements in model interpretability, support for
emerging hardware architectures, and advancements in automated machine learning (AutoML),
making TensorFlow a dynamic and forward-looking library.

In conclusion, TensorFlow has become a cornerstone in the world of deep learning, providing a
robust and versatile platform for building, training, and deploying machine learning models. Its
widespread adoption across various industries and its active community contribute to its ongoing
success as a leading open-source machine learning library. Whether you're a researcher pushing
the boundaries of AI or a developer building intelligent applications, TensorFlow offers the tools
and resources to bring your ideas to life

18
 Chapter-3
Convolutional Neural Networks (CNNs)
Introduction:
Convolutional Neural Networks (CNNs) represent a groundbreaking advancement in the field of
deep learning, specifically tailored for tasks involving visual data such as image recognition,
object detection, and image classification. Developed to mimic the visual processing of the
human brain, CNNs have demonstrated remarkable success in various computer vision
applications.

Core Concepts:
1. Convolutional Layers:
The fundamental building blocks of CNNs are convolutional layers. These layers employ
convolutional operations to extract features from input images. Convolution involves sliding a
filter (also known as a kernel) over the input image, performing element-wise multiplications
and aggregating the results to create a feature map. This process enables the network to
recognize hierarchical patterns and spatial hierarchies in the input data.

2. Pooling Layers:
Pooling layers, often used in conjunction with convolutional layers, reduce the spatial
dimensions of the input volume while retaining essential information. Max pooling, for example,
selects the maximum value from a group of values, thereby preserving the most prominent
features and reducing computational complexity.

3. Fully Connected Layers:

Following convolutional and pooling layers, fully connected layers are employed to make
predictions based on the extracted features. These layers connect every neuron to every neuron
in the preceding and succeeding layers, allowing the network to learn complex relationships and
make high-level abstractions.

19
4. Activation Functions:
Activation functions, such as ReLU (Rectified Linear Unit), introduce non-linearity to the model,
enabling it to learn intricate patterns and relationships within the data. ReLU, for instance,
replaces all negative values in the feature map with zeros, enhancing the network's ability to
capture complex features.

Key Features:
1. Parameter Sharing:
CNNs leverage parameter sharing to reduce the number of learnable parameters in the
network. The same set of weights is applied to different parts of the input data, promoting
feature reuse and making CNNs more efficient and capable of learning hierarchical
representations.

2. Translation Invariance:
Through the use of convolutional operations, CNNs exhibit translation invariance, meaning they
can recognize patterns regardless of their location in the input space. This property is crucial for
tasks like image recognition, where the position of an object within an image may vary.

3. Hierarchical Feature Learning:

CNNs excel in learning hierarchical features. Lower layers typically capture simple features like
edges and textures, while higher layers learn more complex patterns and representations. This
hierarchical approach allows CNNs to understand the composition of objects in a manner similar
to how humans perceive visual information.

Applications:
1. Image Classification:
One of the primary applications of CNNs is image classification. Networks are trained on labeled
datasets to recognize and classify objects or scenes within images with high accuracy.

2. Object Detection:

20
CNNs play a vital role in object detection, identifying and localizing multiple objects within an
image. Region-based CNNs (R-CNNs) and their variants are commonly used for object detection
tasks.

3. Facial Recognition:
CNNs are extensively used in facial recognition systems, enabling the detection and
identification of faces in images and videos. They have applications in security, authentication,
and entertainment.

4. Medical Imaging:
In the field of healthcare, CNNs are applied to tasks such as medical image analysis, aiding in the
detection of anomalies in X-rays, MRIs, and CT scans. They contribute to the diagnosis and
treatment of various medical conditions.

Challenges and Future Directions:

While CNNs have achieved remarkable success, challenges persist, such as the need for large
labeled datasets and susceptibility to adversarial attacks. Future directions include improving
interpretability, addressing bias in training data, and exploring novel architectures to enhance
performance and efficiency.

In conclusion, Convolutional Neural Networks have revolutionized the field of computer vision,
enabling machines to perceive and interpret visual information with human-like accuracy. Their
applications span across diverse domains, from image recognition to medical diagnostics,
making CNNs a cornerstone in the era of artificial intelligence and visual understanding.

21
 CHAPTER 4
PROJECT: BRAIN TUMOR DETECTION USING TRANSFER
LEARNING
4.1Introduction:
Brain tumor detection is a critical area in medical imaging, where the timely and accurate
identification of tumors is essential for effective treatment. Leveraging advanced technologies
such as deep learning, particularly transfer learning, has proven to be a promising approach in
enhancing the accuracy and efficiency of brain tumor detection. Transfer learning involves
utilizing knowledge gained from one task to improve the performance of another, reducing the
need for extensive labeled datasets and training time.

4.2Benefits of Brain Tumor Detection Using Transfer Learning:

1. Improved Performance:
Transfer learning allows the model to leverage pre-trained neural networks on large datasets
from diverse domains. This initial learning enables the model to capture generic features and
patterns, enhancing its ability to recognize intricate details in brain images, leading to improved
performance.

2. Reduced Data Requirements:

Brain tumor datasets are often limited in size due to the challenges associated with obtaining
labeled medical images. Transfer learning mitigates this issue by allowing the model to benefit
from knowledge gained in other domains, reducing the need for an excessively large dataset
specific to brain tumor detection.

3. Faster Training Times:

Training deep neural networks from scratch on medical imaging datasets can be
computationally intensive and time-consuming. Transfer learning accelerates the training
process by initializing the model with weights learned from a different but related task, enabling
quicker convergence and deployment.

4. Adaptability to Limited Resources:

22
Transfer learning makes the development of accurate brain tumor detection models more
feasible, even when computational resources are limited. By building on the knowledge
acquired from pre-trained models, it becomes possible to achieve state-of-the-art results with
less demand on hardware and time.

4.3Dataset for Brain Tumor Detection:

For a brain tumor detection project using transfer learning, the choice of dataset is crucial. An
ideal dataset should encompass a diverse set of brain images, including both normal and tumor
cases. Several publicly available datasets have been curated for medical imaging tasks, and they
can serve as a foundation for training and validating the model. Some widely used datasets for
brain tumor detection include:

1. Brain Tumor Image Segmentation (BRATS):

BRATS provides a collection of multi-modal brain tumor images, including Magnetic Resonance
Imaging (MRI) scans. It includes high-grade and low-grade glioma cases, making it valuable for
training models to distinguish between different tumor types.

2. MICCAI Brain Tumor Segmentation (BraTS) Challenge Data:

The BraTS challenge dataset is designed for evaluating algorithms in brain tumor segmentation.
It includes a large collection of MRI scans with annotations for tumor regions, making it suitable
for training models focused on precise tumor localization.

3. TCIA Glioma Image Data:

The Cancer Imaging Archive (TCIA) hosts a collection of glioma images obtained through various
imaging modalities, providing a resource for developing models that can generalize across
different data acquisition techniques.

4. LGG and HGG Datasets:

Low-grade glioma (LGG) and high-grade glioma (HGG) datasets contain images specifically
categorized based on tumor grade. These datasets are useful for projects aiming to differentiate
between different tumor grades.

23
In conclusion, a brain tumor detection project using transfer learning offers a range of benefits,
including improved performance, reduced data requirements, faster training times, and
adaptability to limited resources. The choice of an appropriate dataset is a crucial step in
ensuring the model's accuracy and generalization capabilities, and leveraging well-curated
medical imaging datasets is essential for the success of the project.

4.4MODEL EVALUATION
The model's evaluation involved testing its performance on an unseen test dataset, where predictions
were generated using random values. This rigorous assessment allowed us to gauge the model's ability
to generalize to new data and determine its effectiveness in making predictions beyond the training
data, ensuring its reliability and robustness in practical applications.

7.9 CODING THE PROJECT IN COLAB

24
25
26
27
Conclusion:

In the realm of medical imaging, particularly in the context of brain tumor detection, the integration of
transfer learning has proven to be a transformative approach. The benefits offered by this methodology,
including improved performance, reduced data requirements, faster training times, and adaptability to
limited resources, address critical challenges in the development of accurate and efficient detection
models.

Transfer learning empowers models to leverage knowledge acquired from unrelated domains, thereby
enhancing their ability to discern intricate patterns and features within brain images. This is particularly
crucial in the medical field, where obtaining large labeled datasets can be challenging. By reducing the
reliance on extensive domain-specific data, transfer learning makes the development of robust brain
tumor detection models more accessible and feasible.

28