Scribd
Upload
8 views12 pages

Projectppt Edited

This document discusses the application of U-Net, a deep learning architecture, for brain tumor segmentation in MRI scans, highlighting its ability to perform precise pixel-level segmentation. The project aims to enhance early detection, reduce manual segmentation errors, and improve diagnosis and treatment planning for brain tumors. Results indicate that U-Net can automate and improve the analysis of medical images, leading to faster and more consistent patient care.

Uploaded by

Keerthi Vundavalli
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
8 views12 pages

Projectppt Edited

This document discusses the application of U-Net, a deep learning architecture, for brain tumor segmentation in MRI scans, highlighting its ability to perform precise pixel-level segmentation. The project aims to enhance early detection, reduce manual segmentation errors, and improve diagnosis and treatment planning for brain tumors. Results indicate that U-Net can automate and improve the analysis of medical images, leading to faster and more consistent patient care.

Uploaded by

Keerthi Vundavalli
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/ 12

BRAIN TUMOUR SEGMENTATION USING

U-NET
by
Keerthi Vundavalli
108121138

In fulfillment of Summer Internship for the award of the degree


of
BACHELOR OF TECHNOLOGY
in
ELECTRONICS AND COMMUNICATION ENGINEERING
Introduction
• Brain tumor segmentation using U-Net is a specialized application of deep learning in medical
imaging, particularly in the detection and delineation of brain tumors in MRI scans.
• The U-Net architecture, a type of convolutional neural network (CNN), is well-suited for this task
due to its ability to perform precise pixel-level segmentation.
• Early detection and accurate segmentation of brain tumors are crucial for effective diagnosis,
treatment planning, and monitoring of patient progress.Manual segmentation by radiologists is
time-consuming and subject to inter-observer variability, making automated methods highly
desirable.
• Automated segmentation provides consistent and objective results, which can enhance the
precision of treatment plans and lead to better patient outcomes. By reducing the time needed
for manual analysis, healthcare providers can quickly develop strategies, improved survival rates
for patients with brain tumors.
Objectives
• Accurate segmentation: The goal is to accurately identify the location and extent of a
brain tumor in a magnetic resonance imaging (MRI) scan. This is done by classifying each
pixel or voxel of the input data into a pre-set tumor region category.
• Early detection: Deep learning can help identify brain tumors early, which can lead to
preventative measures to save lives
• Reduced error: Deep learning can reduce the cost, time, and error of manual
segmentation
• Improved diagnosis and treatment: Brain tumor segmentation can help clinicians
diagnose and treat patients
• Case-based reasoning: A database of cases from multiple views and imaging modalities
can help medical experts monitor cases and respond to new cases
Methodology
Dataset :
The dataset referred is for brain tumor segmentation, available on Kaggle. It contains MRI images of
the brain with labeled regions indicating the presence of tumors.

https://www.kaggle.com/datasets/nikhilroxtomar/brain-tumor-segmentation
Model architecture
U-Net, an image is converted into a vector and then the same mapping is used to convert it again to
an image. This reduce Using a combination of convolution, up-convolution, pooling and skip
connections, the U-Net is able to extract and capture complex features, while also keeping and
reconstructing spatial information the distortion by preserving the original structure of the image.

• Encoder: The network begins with a series of convolutional layers that downsample the input
image, extracting features and capturing context. This part is similar to the encoder in a typical
convolutional neural network (CNN).

• Bottleneck: At the bottom of the U-shape, the bottleneck consists of two convolutional layers
that further process the features extracted by the encoder. This stage serves as a bridge
between the encoder and decoder, retaining the most critical information while discarding the
less important details.
• Decoder: The decoder mirrors the encoder but uses upsampling layers instead of pooling layers.
Each upsampling step is followed by a concatenation with the corresponding feature map from the
encoder. This step ensures that fine-grained information lost during downsampling is recovered.
Convolutional layers are applied after concatenation to refine the segmentation map.

• Skip Connections: One of the key features of U-Net is the use of skip connections between
corresponding layers in the encoder and decoder. These connections concatenate features from
the encoder with those in the decoder, helping to preserve spatial details and improve
segmentation accuracy.

• Output Layer: The final layer of the U-Net is a 1x1 convolution that reduces the number of output
channels to the number of classes, typically followed by a softmax or sigmoid activation function
for multi-class or binary segmentation, respectively.
Block diagram of the UNET
architecture
Training
Data splitting divides a dataset into three main subsets:Training Set: The model is trained on this
data, learning patterns to make predictions.Validation Set: During training, the model is evaluated
on this data to fine-tune parameters and avoid overfitting, ensuring it generalizes well to new
data.Testing Set: After training and validation, the model is tested on this data to evaluate its
performance, providing an estimate of how it will perform in real-world scenarios.
• Training split:
Training data : 60%
Validation data : 20%
Test data : 20%
• Hyperparameters:
batch size = 5
learning rate = 1e-4
number of epochs = 10
• Framework: TensorFlow and Keras.
Results
• Evaluation metrics
The model was evaluated using the test set, and the following average scores were obtained:
F1: 0.941484716
Jaccard: 0.889438944
Recall: 0.990808824
Precision: 0.896838602
Visual results

input image mask image predicted image


Conclusion
This project successfully used the U-Net model to accurately identify and segment brain tumors in
MRI scans. The results show that deep learning models like U-Net can help automate and improve
the process of analyzing medical images. By reducing the time and variability associated with
manual segmentation by radiologists, these models can assist in faster and more consistent
diagnosis and treatment planning.
Automated segmentation tools like this can enhance the overall quality of care by providing quick
and reliable results. Adjusting the learning rate, batch size, and number of epochs can significantly
impact the performance of a U-Net model for brain tumor segmentation.
References
• https://www.sciencedirect.com/science/article/pii/S2772442522000429#d1e630

• https://ieeexplore.ieee.org/document/9827132

• https://www.sciencedirect.com/science/article/pii/S1746809422008813

You might also like