0% found this document useful (0 votes)
11 views28 pages

Batch 27

The document presents a project on automating breast cancer diagnosis using deep learning techniques applied to ultrasound imaging. It outlines the methodology involving data collection, preprocessing, feature extraction using transfer learning models, and classification with machine learning algorithms, emphasizing the use of ensemble learning for improved accuracy. The project aims to enhance diagnostic efficiency and reduce reliance on manual interpretations by radiologists.

Uploaded by

Himabindu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
11 views28 pages

Batch 27

The document presents a project on automating breast cancer diagnosis using deep learning techniques applied to ultrasound imaging. It outlines the methodology involving data collection, preprocessing, feature extraction using transfer learning models, and classification with machine learning algorithms, emphasizing the use of ensemble learning for improved accuracy. The project aims to enhance diagnostic efficiency and reduce reliance on manual interpretations by radiologists.

Uploaded by

Himabindu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 28

ADITYA INSTITUTE OF TECHNOLOGY AND MANAGEMENT

DEPARTMENT OF COMPUTER SCIENCE AND MANAGEMENT


(Approved by AICTE, Permanently Affiliated to JNTUGV, Vizianagaram,
Accredited by NBA & NAAC with A+)
(AUTONOMOUS)
K.KOTTURU, TEKKALI - 532201
BREAST CANCER CATEGORIZATION USING

DEEP LEARNING

PRESENTED BY:

21A51A05B8 BONGI SAJANI

21A51A0581 INUMURTHY AKHILA


Under the guidance of:
22A55A0507 GALI BHAVANI PRASAD
Sri T Chalapathi Rao
(Asst. Professor ) 21A51A05A7 SAVARA NIRMALA

BATCH 27 2
TABLE OF CONTENTS

 ABSTRACT
 INTRODUCTION
 PROBLEM DEFINITION
 LITERATURE SURVEY
 METHODOLOGY
 RESULT ANALYSIS
 PLAN OF ACTION

20X 3
X
ABSTRACT
Breast cancer remains a significant global health concern, emphasizing the need for early and accurate
diagnosis to improve patient outcomes. This project proposes an intelligent system aimed at automating breast
cancer identification using ultrasound imaging. The system begins with the collection of a diverse dataset
comprising ultrasound images categorizing breast conditions into normal, benign, and malignant cases.
Employing robust pre-processing techniques such as data augmentation, cropping, and resizing standardizes the
images for subsequent analysis.
Transfer learning models— VGG-19, MobileNet, and InceptionNet —are utilized to extract meaningful features
from the ultrasound images, enhancing the efficiency of model training and evaluation. The dataset undergoes
rigorous evaluation using 10-fold cross-validation, ensuring robust model performance. Machine learning
models including Support Vector Machines (SVM), Recurrent Neural Networks (RNN), and Random Forests
(RF) are employed for breast cancer classification, fine-tuned to distinguish between benign and malignant
tumors. The UBC Benchmark dataset, comprising 780 images annotated with benign, malignant, and normal
labels, serves as a critical resource for algorithm development and evaluation. This project underscores the
potential of ensemble transfer learning strategies in advancing breast cancer diagnosis through intelligent
ultrasound imaging

20X 4
X
INTRODUCTION

 Breast cancer is a disease where cancerous cells form in the tissues of the breast. It
typically begins in the cells of the milk-producing ducts or the glandular tissues.
 Breast cancer tumors can be malignant or benign. Malignant tumors are cancerous, capable
of spreading (metastasizing) to other body parts, and are potentially life-threatening. Benign
tumors are non-cancerous, usually do not spread, and are generally less dangerous, but
they may still require monitoring or removal.
 Breast cancer affects approximately 2 million women globally each year, causing around
685,000 deaths. About 1 in 8 women will develop breast cancer in their lifetime. Early
detection has led to a 99% 5-year survival rate for localized cases, though the risk increases
with age, particularly after 50.

5
INTRODUCTION(Continued)

 Common symptoms include a lump in the breast, changes in breast shape, dimpling
of the skin, discharge from nipple.
 There are some traditional machine learning approaches like Logistic Regression,
Decision Trees, Random Forests , K-Nearest Neighbors ,Naïve Bayes to classify
the breast tumors .
 So, To try to increase the accuracy, precision, Recall and F1-score of classification
to detect the type of tumors, we are using the machine learning techniques.
PROBLEM DEFINITION

This project aims to improve the diagnosis of breast cancer by automating the
analysis of ultrasound images. It addresses the variability and subjectivity of
manual interpretations by radiologists. It applies deep learning and transfer
learning techniques, using models such as VGG-19, MobileNet, and
InceptionNet for feature extraction. The dataset is rigorously evaluated using
10-fold cross-validation to minimize overfitting and validate the model's
generalization capabilities. Such as SVM, RNN, RF are used machine
learning models classify breast tissue for normal, benign,
and malign. Moreover, a user interface
of HTML generates complete reports to support the diagnostics on
classification outcomes.

7
LITERATURE SURVEY
S.No Author Title Year Techniques Limitations
used
1 Rohit Classification of mammogram Model’s reliance
AgrawalNavneet breast cancer enhancement,dee on discrete
Pratap Singh.et.al from p convolutional cosine
mammograms 2024 neural transformation
using enhance network,discrete may lead to
image reduction cosine transform sensitivity
and deep towards image
learning preprocessing
framework variations.
2 Maged Deep learning Convultional It covers various
Nasser.et.al based methods Neural Network techniques lacks
for breast cancer discussion on
diagnosis: A 2023 real-world
Systemic review challenges of
implementing
these methods
such as data
privacy issues.

3 Tariq Mahmood, Breast lesions Random Despite strong


Jianqiang Li.et.al classifications of forest,Root Mean performance
mammographic Square, Support metrics the cnn
images using a 2022 Vector Machine model lacks
deep explanability.
convolutional
neural network-
S.No Author Title Year Techniques Limitations
used

4 Muhammad Breast Cancer Convolutional Despite achieving high


Shahid Iqbal.et.al Dataset, Neural Network, performance DL
Classification and Artificial Neural models lack model
Detection Using 2022 Network,Google interpretability in
Deep Learning LeNet complex architectures
like GoogleLeNet

5 Viswanatha Breast cancer Support Vector It lacks cross-modality


Reddy detection based on Machine,Random generalization,as it is
Allugunti.et.al thermographic Forest,Convoluti limited to
images using 2022 onal Neural thermographic images.
machine learning Network
and deep learning
algorithms

6 Basem S Breast cancer Magnetic It lacks robustness in


Abunasser.et.al detection and Resonance handling complex
classification Imaging,compute imaging variations
using deep 2022 d commonly found in
learning Xception tomography,deep real-world computed
algorithm learning tomography scans.
Xception
algorithm
S.No Author Title Year Techniques Limitations
used

7 Rayees Ahmad Breast cancer Mammography, Using these


Dar.et.al detection using Histopathology, techniques lacks
deep learning: 2022 Ultrasound robustness and
Datasets, generalizability
methods, and across diverse
challenges datasets
ahead

8 Sharmin Ara, Malignant and Support Vector Lack


Annesha benign breast Machine, explainability in
Das.et.al cancer 2021 Logistic terms of feature
classification Regression, K- importance and
using machine Nearest decision
learning Neighbors, boundaries
algorithms Decision Tree, particularly in
Naive Bayes black-box
and Random models.
Forest

9 Muhammad comprehensive Artificial Neural It emphasizes


Firo survey on deep- Network,Deep technical
Mridha.et.al learning-based 2021 Belief performance but
breast cancer Network,Convol lacks mention of
diagnosis utional Neural real-world
Network clinical
integration,usab
ility or
validation.
METHODOLOGY

As a part of the classification step of our breast cancer


ultrasound image analysis, an ensemble learning
approach is utilized to improve diagnostic accuracy. After
feature extraction with advanced transfer learning
models, including VGG-19, MobileNet, and
InceptionNet, we then proceed with training classifiers
such as Support Vector Machine (SVM), Recurrent
Neural Networks (RNN), and Random Forest (RF). Each
model is further fine-tuned for optimal performance, with
SVM exploring various kernel functions, RNN capturing
spatial dependencies, and RF leveraging multiple
decision trees to improve robustness.
METHODOLOGY(continued)

We then implement a stacking ensemble technique in


order to combine the outputs of these classifiers, thereby
maximizing predictive performance while minimizing
individual weaknesses. The system is rigorously
evaluated using 10-fold cross-validation, ensuring
reliability and generalizability of results, with
performance metrics such as accuracy, precision, recall,
F1-score, and AUC computed to assess effectiveness.
This methodology aims to automate breast cancer
diagnosis, reducing reliance on manual interpretation and
improving diagnostic speed and consistency.
Ensemble Learning
An ensemble learning approach, which aggregates Support Vector
Machine (SVM), Recurrent Neural Networks (RNN), and Random Forest
(RF), is applied to enhance classification performance. Variability is
reduced and accuracy is improved to produce robust and reliable results
in the clinical setting.
Evaluation Using 10-Fold Cross-Validation
The system undergoes rigorous evaluation through 10-fold cross-
validation. In this approach, the dataset is split into 10 parts, and the
model is trained on 9 parts while being tested on the remaining one part.
This process is repeated 10 times, with each part used as a test set once.
The average performance across all folds is calculated, ensuring that the
system performs consistently and avoids overfitting.
WORKING PROCESS:
 Dataset Collection:
Gather a comprehensive Ultrasound Breast Image Cancer (UBC)
Benchmark dataset containing labeled ultrasound images of normal, benign,
and malignant breast tissues.
 Data Preprocessing:
Apply preprocessing techniques such as data augmentation (rotation,
flipping, etc.), cropping, and resizing to enhance image quality and increase
dataset diversity.
Normalize pixel values to ensure consistency across images.
 Feature Extraction:
Utilize pre-trained convolutional neural networks (CNNs) such as VGG-19,
MobileNet, and InceptionNet.Fine-tune these models on the UBC dataset to
adapt them to the specific characteristics of breast ultrasound images.
Extract informative features that capture essential patterns and variations in
the images.
 Feature Storage:
Store the extracted features in a structured format for subsequent analysis, ensuring they are ready for classification.
 Model Selection:
Choose classification models, including Support Vector Machine (SVM), Recurrent Neural Networks (RNN), and
Random Forest (RF).
Fine-tune the hyperparameters of each model to optimize performance.
 Ensemble Learning:
Implement a stacking ensemble technique to combine the predictions of the individual classifiers (SVM, RNN, RF).
Aggregate the outputs of these models using a linear aggregation method to derive a final classification decision.
 Model Evaluation:
Conduct 10-fold cross-validation to evaluate the models on unseen data, ensuring reliability and generalizability.
Compute performance metrics such as accuracy, precision, recall, F1-score, and AUC to assess model effectiveness.
Source Code
Pseudo Code
Result Analysis

Classification Report
Confusion Matrix
OUTPUT SCREEN

INDEX Page: The homepage features a clean and


modern design with a focus on "Ensemble Transfer
Learning Strategies." It invites users to either register or
log in, setting the stage for further interaction with the
application.
Registration Page:

This page allows new users to create an account by


providing their username, email ID, password, and
confirming the password. It is designed for ease of
use, ensuring a smooth onboarding process.
Login Page:

The login page provides fields for users to enter their email
ID and password. This page is essential for returning users to
access their accounts and utilize the diagnostic features of the
application.
HOME PAGE: This page introduces users to the deep learning
techniques employed in the application for detecting breast
cancer. It emphasizes the system's purpose and sets the context
for the diagnostic process.

ABOUT PAGE: The "About" page describes the project's


goals, highlighting its aim to enhance breast cancer diagnosis
through advanced machine learning techniques. It explains
the application’s purpose and potential impact on patient
outcomes.
Users can upload ultrasound breast images for diagnosis on this After uploading an image, this page displays the prediction
page. It includes a file chooser and a submit button, enabling result indicating that the uploaded ultrasound image is classified
seamless interaction for submitting images for analysis. as benign. It reassures users about their diagnosis.
PLAN OF ACTION

8
NPTEL REGISTRATION MAILS

BONGI SAJANI INUMURTHY AKHILA


GALI BHAVANI PRASAD SAVARA NIRMALA
THANK YOU
~Batch
27

20X 9
X

You might also like