Deep Learning for Tumor Detection

Minor Project Submitted in Partial Fulfillment of the Requirements for the Degree
of Bachelor of Technology in the field of Computer Science & Technology

BY
Ayan Mondal 123231110208
Abir Pramanik 123231110201
Sourav Mahanta 123231110218

Under the supervision of

Dr.Sitanath Biswas
Guide Name
Dr.Sitanath Biswas
And co-supervision of
Dr.Sayan Chakraborty
Co-guide name
Dr.Sayan Chakraborty

Department of Computer Science & Technology

JIS College of Engineering

Block-A, Phase-III, Kalyani, Nadia, Pin-741235

West Bengal, India
Dec, 2024
 CERTIFICATE
This is to certify that Ayan Mondal-123231110208,Abir Pramanik-
123231110201, Sourav Mahanta-123231110218 has completed his project entitled
Deep Learning for Tumor Detection, under the guidance of Dr.Sitanath Biswas
in partial fulfillment of the requirements for the award of the Bachelor of
Technology in Computer Science & Technology from JIS college of Engineering
(An Autonomous Institute) is an authentic record of their own work carried out
during the academic year 2024-25 and to the best of our knowledge, this work has
not been submitted elsewhere as part of the process of obtaining a degree, diploma,
fellowship or any other similar title.

---------------------------------

Signature of the Supervisor

---------------------------------

Signature of the Co-Supervisor

Place:

Date:

ii
 ACKNOWLEDGEMENT

The analysis of the project work wishes to express our gratitude to Dr.Sitanath
Biswas for allowing the degree attitude and providing effective guidance in
development of this project work. His conscription of the topic and all the helpful
hints, he provided, contributed greatly to successful development of this work,
without being pedagogic and overbearing influence.

We also express our sincere gratitude to Dr.Sitanath Biswas, Head of the

Department of Computer Science & Technology of JIS College of Engineering and
all the respected faculty members of Department of CST for giving the scope of
successfully carrying out the project work.

Finally, we take this opportunity to thank to Prof. (Dr.) Partha Sarkar, Principal of
JIS College of Engineering for giving us the scope of carrying out the project work.

Date:
…………………………………….………………………………….
Ayan Mondal
B.TECH in Computer Science & Technology
3rd Year/5th Semester
Univ Roll--123231110208

…………………………………….………………………………….
Abir Pramanik
B.TECH in Computer Science & Technology
3rd Year/5th Semester
Univ Roll--1123231110201

…………………………………….………………………………….
Sourav Mahanta
B.TECH in Computer Science & Technology
3rd Year/5th Semester
Univ Roll--123231110207

iii
 CONTENTS

Title page
Certificate
Acknowledgement

Abstract
1. Introduction
2. Literature Survey
3. Methodology
4. Proposed Method
5. Result and Discussion
6. Conclusion
Reference
PUBLICATIONS FROM THE WORK

iv
1.Introduction

The human brain is regarded to be one of the most essential organs since it is
responsible for a large number of the body's regulatory processes, including
memory, emotions, vision, motor skills, responses, and breathing. In the event that
a tumour begins to form inside the brain, these functions will be significantly
disrupted [1,3]. This tumour is either a primary brain tumour (BT), which develops
from inside the brain itself and represents the development of brain
tissues themselves, or it is a metastatic BT, which develops in another part of the
body and eventually spreads to the brain. When compared to tumors that originate
in any other organ of the human body, those that occur in the brain provide a
significant diagnostic challenge. Because the brain has the “Blood-Brain Barrier
(BBB)”, ordinary radioactive markers are unable to detect the hyperactivity of
tumour cells in the body [2]. Consequently, MRI scans are considered to be the most
effective diagnostic tracers for detecting breaches in the BBB. Fig. 1(A) and (B)
depicts the images of healthy brain and tumour brain.

v
2.Literature Survey

High Accuracy and Sensitivity

• Deep learning models, especially convolutional neural networks (CNNs),
excel at recognizing intricate patterns in medical imaging data, often
surpassing traditional machine learning methods and even human experts.
• Early and accurate detection of cancer improves prognosis, treatment
planning, and patient outcomes.
Automation and Scalability
• Deep learning enables automated analysis of medical data (e.g., radiology
images, pathology slides, or genomic data), reducing the workload of
radiologists and pathologists.
• Scalable models can process thousands of samples in a fraction of the time
required for manual analysis.
Handling Complex Data
• Tumor detection often involves multi-dimensional, heterogeneous data:
o Medical Images: CT, MRI, mammograms, histopathology slides.
o Genomics and Multi-Omics: Deep learning models can analyze
sequencing data to identify biomarkers.
o Clinical Text: Models like transformers can extract insights from
patient histories or reports.
• Deep learning effectively integrates such data to improve diagnostic accuracy.
Improved Feature Extraction
• Deep learning automatically learns hierarchical and high-level features
from raw data, enabling the detection of subtle abnormalities that might be
invisible to human eyes.

vi
3.Methodology

Classification and Object Detection for Brain Tumor Detection -

Deep learning techniques for brain tumor detection follow a similar methodological
approach as used in other imaging tasks, leveraging advanced neural networks for
classification and object detection.

Classification Models -
Classification models analyze brain MRI scans to categorize them under labeled
classes, such as normal or tumor-affected. Commonly used neural networks for
image classification in brain tumor detection include Residual Networks (ResNet),
Densely Connected Convolutional Networks (DenseNet), MobileNets, EfficientNet,
and ConvNeXt. These models extract hierarchical features from the image, enabling
accurate differentiation between healthy brain tissue and tumor-affected regions.

Object Detection Models-

Object detection models build upon classification by not only determining the
presence of a tumor but also pinpointing its approximate location within the MRI
scan. These models typically output bounding boxes around suspected tumor regions,
aiding in interpretability and facilitating clinical decision-making. Prominent neural
networks used for object detection include You-Only-Look-Once (YOLO), Region-
Based Convolutional Neural Networks (R-CNN), and Single-Shot Multibox
Detection (SSD).

Applications in Brain Tumor Detection-

Classification and object detection have been successfully applied to tasks involving
the analysis of brain MRI scans. These models automate the detection of tumor
presence and type (e.g., glioma, meningioma, or pituitary tumor), assisting

vii
radiologists in prioritizing cases for detailed review. They also enable real-time
assessments in emergency settings where quick decisions are critical.
Furthermore, classification models have been extended to predict tumor grade,
prognosis, and treatment response, aiding personalized medical interventions. These
models are particularly valuable in identifying subtle differences in tumor
characteristics, such as histopathological features, by leveraging deep learning-based
radiomics.
Simplified CNN Architecture for Brain Tumor Classification
The architecture of a convolutional neural network (CNN) for brain tumor detection
involves blocks of convolutional layers that extract meaningful features, such as
edges, textures, and patterns, from the MRI images. These features are fed into fully
connected layers for final prediction, classifying the image into categories like
"tumor" or "normal."

Performance Evaluation-
The performance of classification and object detection models in brain tumor
detection is assessed using metrics such as Area Under the Receiver Operating
Characteristic Curve (AUC) and confusion matrix-derived metrics (e.g., accuracy,
precision, recall, and F1-score).
• AUC: Measures the model's ability to distinguish between positive (tumor
presence) and negative (normal) classes. An AUC of 1.0 indicates perfect
classification, while 0.5 signifies performance equivalent to random guessing.
• Comparison with Machine Learning Models: The performance of deep
learning classifiers is compared against traditional machine learning models
such as XGBoost, support vector machines (SVM), random forests, logistic
regression, naive Bayes, stochastic gradient descent, and neural networks.

viii
4. Proposed Method

The proposed methodology leverages a CNN-based architecture to detect brain tumors from MRI
images. The dataset was preprocessed by resizing images to a uniform size of 236x236 pixels
and normalizing pixel values to improve model training. Data augmentation techniques such as
rotation and flipping were applied to address overfitting.

The CNN architecture comprises multiple convolutional layers followed by max-pooling layers
and fully connected layers.

The model was trained using the using metrics such as accuracy, precision, and recall. And was
applied to interpret the results and validate that the model focused on tumor regions during
prediction.

Here are some of the skimages of our programs.

ix
Performance Evaluation Metrics

The performance of deep learning algorithms for brain tumor detection was evaluated
and compared against machine learning models using metrics derived from the
confusion matrix. These metrics include:
• True Positive Rate (TPR): Also known as sensitivity, it measures the
proportion of actual positive cases (presence of brain tumor) correctly
identified by the model. It is calculated as the ratio of true positives to the sum
of true positives and false negatives. High sensitivity is critical for reducing
missed diagnoses of brain tumors.
• True Negative Rate (TNR): Also known as specificity, it measures the
proportion of actual negative cases (absence of brain tumor) correctly
identified. It is calculated as the ratio of true negatives to the sum of true
negatives and false positives. High specificity ensures minimal
misclassification of healthy patients as having a tumor.
• False Negative Rate (FNR): This measures the proportion of actual positive
cases incorrectly classified as negative. It is calculated as the ratio of false
negatives to the sum of true positives and false negatives. In brain tumor
detection, a low FNR is crucial to minimize missed tumor cases, as this could
lead to delayed diagnoses.
• False Positive Rate (FPR): This quantifies the proportion of actual negative
cases incorrectly classified as positive. It is calculated as the ratio of false
positives to the sum of true negatives and false positives.
• False Detection Rate (FDR): This measures the proportion of false positives
among all predicted positives. It is calculated as the ratio of false positives to
the sum of true positives and false positives.
• false negatives to the sum of true negatives and false negatives. A low FOR is
vital in brain tumor detection to ensure minimal missed diagnoses.

x
Cross-Validation and Experimental Setup-

The performance metrics were evaluated through Monte Carlo experiments using a
k-fold cross-validation approach. This involves splitting the dataset into kkk training
and testing subsets. The classifiers, both machine learning and deep learning, are
trained on the training subset and tested on the held-out testing subset to assess
predictive accuracy and robustness.

Dataset and Challenges-

Brain tumor detection requires high-quality datasets containing MRI images
validated and labeled by radiologists as ground truth. Publicly available datasets, such
as the Brain Tumor Image Segmentation (BraTS) dataset, provide valuable resources
for training and testing deep learning models. BraTS includes multimodal MRI scans
(T1, T2, FLAIR, and T1Gd) with segmentation masks outlining tumor regions,
enabling both classification and object detection tasks.
Despite the availability of large datasets, challenges persist:
• Heterogeneity in Imaging: Variations in imaging protocols, devices, and
demographics between datasets can affect model performance when applied to
external datasets or new facilities.
• Overfitting to Specific Datasets: Models trained exclusively on data from a
single institution may struggle to generalize to other settings.
To address these issues, transfer learning is employed. Models pre-trained on large
public datasets are fine-tuned on smaller, institution-specific datasets to adapt to local
conditions. Persistent monitoring and periodic retraining ensure sustained
performance in diverse settings.

xi
Neural Network Training-

The development of a deep learning model involves selecting an appropriate neural

network architecture and curating a robust training dataset. During training, the
neural network extracts and learns relevant features from MRI scans, using labeled
ground truth data to optimize its performance. Weighted connections between
neurons are adjusted iteratively to improve accuracy, enabling the model to make
predictions autonomously on new data. Recent advancements in deep learning have
simplified the creation of performant models through general-purpose architectures
and self-configuring pipelines. Networks such as ResNet, DenseNet, and U-Net have
proven effective in brain tumor detection by handling both classification and
segmentation tasks with high accuracy.

xii
5.Result and Discussion

Comparison to Other Diseases-

Brain tumors, like other forms of cancer, result from abnormal cellular growth
forming neoplasms in brain tissues. Cancer remains the second leading cause of
death worldwide, with approximately 9.6 million fatalities in 2018. Brain tumors
represent a critical subset of cancer types due to their location and impact on
neurological functions. The early diagnosis of brain tumors significantly increases
the chances of effective treatment and long-term survival. However, diagnosis at
advanced stages often results in poor outcomes, making early detection vital.
Medical imaging modalities such as MRI, CT, and PET are integral to brain
tumor detection, providing critical insights into tumor location, size, and
progression. The manual interpretation of these images is challenging due to their
complexity and the high volume of data, which increases the likelihood of human
error. To address these challenges, automated systems using computer-aided
diagnosis (CAD) have been developed since the 1980s. The integration of deep
learning techniques has further revolutionized CAD, enabling more accurate and
efficient tumor detection and diagnosis.
Deep learning methods analyze medical images to detect patterns indicative of
tumors, helping healthcare professionals identify brain tumors at earlier stages.
These methods are categorized based on their learning strategies: supervised
learning, semi-supervised learning, unsupervised learning, and reinforcement
learning. Supervised learning is most commonly used for brain tumor detection,
relying on labeled datasets to train models to classify tumors accurately.
Tumor Detection Process
The process of detecting brain tumors using medical imaging and deep learning
involves four main phases: pre-processing, segmentation, post-processing, and
classification.

Pre-Processing Phase-
Pre-processing prepares raw medical images for analysis by removing noise and
enhancing image quality. Brain MRI and CT scans often contain noise such as
Gaussian or speckle noise, which can obscure critical details. Pre-processing steps
such as contrast enhancement, skull stripping, and grayscale conversion
improve the clarity and accuracy of images for further analysis.
• Noise Removal: Techniques like Gaussian filters, adaptive median filters,
and mean filters reduce noise.
• Contrast Adjustment: Enhances visibility between brain tissues and
tumors.

xiii
 • Skull Stripping: Removes non-brain elements to focus on tumor regions.
These steps are crucial for reducing artifacts and highlighting the tumor
regions within the brain.
Image Segmentation Phase
Segmentation divides medical images into meaningful regions, isolating the tumor
from surrounding brain tissues. This phase is essential for identifying the tumor’s
location and boundaries.
Segmentation methods for brain tumors include:
1. Threshold-Based Methods: Identify regions based on intensity thresholds
(e.g., Otsu’s method).
2. Region-Based Methods: Use algorithms like watershed segmentation and
region growing to delineate tumor boundaries.
3. Pixel-Based Methods: Leverage machine learning models like artificial
neural networks and fuzzy clustering to classify each pixel.
4. Model-Based Methods: Employ deformable models to adapt to irregular
tumor shapes.
Hybrid segmentation approaches combining multiple techniques often yield
more accurate results.

Post-Processing Phase-
Post-processing refines segmented images to enhance tumor detection accuracy.
Methods like region merging, smoothing, and border expansion improve the
quality of segmentation. Feature extraction techniques such as principal
component analysis (PCA) and grey-level co-occurrence matrix (GLCM)
identify critical tumor characteristics for further analysis. These features include
tumor shape, size, and texture, which are essential for accurate classification.
Classification Phase
The classification phase involves applying deep learning models to identify tumor
types based on extracted features. Commonly used models include:
• Convolutional Neural Networks (CNNs): Excels in spatial data analysis
and tumor classification.
• Recurrent Neural Networks (RNNs): Analyze sequential data and time-
evolution of tumor growth.
• Hybrid Models: Combine CNNs and long short-term memory (LSTM)
networks for enhanced accuracy.
Deep learning models are trained on labeled datasets to classify brain tumors (e.g.,
gliomas, meningiomas) with high precision. After sufficient training, these models
predict tumor types for new images, achieving significant accuracy in clinical and
experimental studies.

xiv
Classification of Deep Learning Models:-

The classification phase involves the application of deep learning methods for
classifying medical images into respective cancer categories based on the features
extracted by the post-processing phase as described in Sub-Section 2.3. Different
types of deep learning methods have been developed based on different principles.
Broadly, deep learning methods can be divided into four categories described in
Section 3. The classification phase involves training and testing the deep learning
classifier for detecting cancer types based on the extracted features from images.
The first phase trains a deep learning classifier using training dataset images. After
appropriate training iterations, the trained deep learning model predicts the cancer
type of unknown images based on the extracted features. The most commonly used
deep learning methods for cancer detection are convolution neural network (CNN),
long short term memory (LSTM), recurrent neural network (RNN), and gated
recurrent units (GRU). 3 Classification of Deep Learning Models Correct acquisition
of medical images and their interpretation plays a significant role in accurately
detecting and diagnosing cancer disease. There exist many image capturing devices
with high-resolution such as CT scan, MRI scan and X-ray scans. After their pre-
processing phase, the disease detection system involves extracting relevant features
from these medical images and training the models from the extracted features. The
trained model is further used to detect the disease from respective unknown medical
images.
Due to considerable variation in medical images of different patients, the
conventional machine learning method cannot provide authentic and accurate
results. In recent years, deep learning methods have been successfully employed in
different fields, specifically in analyzing medical images. These methods are
beneficial and efficient in analyzing medical images to detect diseases.

xv
Deep learning methods are a subset of machine learning methods that allows
approximating the outcome using the given data set and train the model as per the
result. Deep learning methods involve neural networks with multiple layers of
neurons like the input layer, multiple hidden layers, and output layer. Due to the
presence of multiple layers, the deep learning model gets trained more accurately.
The deep learning models can be categorized based upon their learning strategies
into four classes, namely, supervised learning, semi-supervised learning,
unsupervised learning and reinforced learning models.

Here are Some of the output images -

xvi
6.Conclusion

In conclusion, our project on deep learning for brain tumor detection successfully
demonstrated how modern machine learning techniques can be applied to classify
medical images and assist in early cancer diagnosis. By utilizing convolutional
neural networks (CNNs), we achieved promising accuracy in identifying cancerous
and non-tumor from a sample dataset. This project highlighted the importance of
data preprocessing, model optimization, and evaluation in building effective AI
solutions. While our model shows potential, further work is needed to test it on
larger, real-world datasets and improve its generalizability. Overall, this project
provided valuable insights into the application of AI in healthcare and strengthened
our understanding of deep learning techniques. This experience not only deepened
our understanding of deep learning techniques but also highlighted the potential and
responsibility of using AI for impactful solutions like early cancer detection, offering
hope for improved diagnostic accuracy and patient care in the future.

xvii
Reference

1. https://genomemedicine.biomedcentral.com/articles/10.1186/s13073-021-00968
2. https://pmc.ncbi.nlm.nih.gov/articles/PMC10377683/
3. https://www.mdpi.com/2075-4418/14/8/848
4. https://bvmengineering.ac.in/NAAC/Criteria1/1.3/1.3.4/18CP814_Thesis.pdf

xviii
Publications From The Work

xix