Scribd
Upload
9 views

ultrasound nerve segmentation

Uploaded by

rahimango6007
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
9 views

ultrasound nerve segmentation

Uploaded by

rahimango6007
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 13

Ultrasound

Nerve
Segmentation
Mohammed Mubashir Uddin
Syed Haji Ali
Mir Rahimuddin
Introduction

To identify nerve structures an ultrasound image of the region containing nerves is


taken but identification is a non-trivial task even for the experts in the field as the
image generated by ultrasound are affected by echo perturbations and speckle
noise.
In this project, we will see how we can take advantage of cutting-edge deep
learning techniques to build an end-to-end system where a person would just be
feeding the ultrasound image of the region and a deep learning model will segment
the nerve present in the image.
Use Cases
Nerve damage assessment: Segmenting nerves helps evaluate the extent
and location of nerve damage.
Guiding nerve blocks and injections: Accurate segmentation aids in the
precise administration of nerve blocks and injections.
Preoperative planning for nerve-related surgeries:
Implementation
To implement the ultrasound nerve segmentation model, you would need a
dataset of ultrasound images with corresponding annotated nerve
segmentations. The dataset is used for training the model. You can utilize
deep learning frameworks such as TensorFlow or PyTorch to implement the
U-Net architecture. The code you provided represents a part of the model
architecture.
DATASET

For this case study, we have collected the data from kaggle

We are provided with the following data:


Train folder: It contains ultrasound images and masks of the
Brachial plexus. A large training set of images where the nerve
has been manually annotated by humans was provided.

Test data: It contains only ultrasound images without mask


.
Network Artitechture

Being an image segmentation problem , wherein, just classifying the


image wouldn't solve it , but segmenting within the image should help.
Hence, upsampling of CNN output has to be done to produce probability
mask. Hence, U-Net can be used.
U-Net has two paths - Contraction Path and Expansion Path.
Contraction path extracts context of the image and doing so, the image
is down sampled.
Then, the Expansion path upsamples the image and outputs a
probability mask of same size as that of input image.
SIMPLER
U- NET ARTITECHTURE UNDERSTANDING

Convolutional layer

Picture
U-NET Artiitechture
Sampling and
unsampling

pooling layer
U- NET ARTITECHTURE

A U-Net architecture, a popular neural network architecture for image


segmentation tasks. It consists of convolutional layers and pooling layers
for downsampling and upsampling the input image. Dropout layers are used
for regularization to prevent overfitting. The code defines a function U_Net
that constructs the U-Net model based on the provided input image tensor
and the number of filters. The model is built using the Keras framework.
U- NET ARTITECHTURE

1. Convolutional Layers: Convolutional layers are the fundamental building


blocks of convolutional neural networks (CNNs). They apply a set of
learnable filters (also known as kernels) to the input image to extract
relevant features. Each filter performs a convolution operation by
sliding over the input image, computing dot products with the
corresponding image patches, and producing feature maps.
Convolutional layers help the model learn hierarchical representations
of the input image.
U- NET ARTITECHTURE

Pooling Layers: Pooling layers are used for downsampling the feature maps
produced by the convolutional layers. The most common type of pooling is
max pooling, where a pooling window slides over the feature map and
selects the maximum value within each window. Max pooling reduces the
spatial dimensions of the feature maps, effectively downsampling them
while preserving the most salient features. Downsampling helps in
capturing large-scale patterns and reducing the computational complexity
of the network.
U- NET ARTITECHTURE

So, in simpler terms, the U-Net architecture is like a computer program that
looks at a picture, finds important features using special filters, and then
makes the picture smaller and bigger again to find and highlight nerves in an
ultrasound image.
Output
The model takes an ultrasound image as input and produces a corresponding
segmentation map, where the pixel values indicate the probability or confidence
of each pixel belonging to a nerve region. The output can be visualized by
overlaying the segmentation map on the original ultrasound image, highlighting
the nerve regions.

The model is trained for 50 epochs with Early stopping , in case the
validation loss doesn't lower further.
An accuracy of around 98% is achieved by the model on the test data set
Output

To-do List
Build a model and train it
Evaluate the model
Building a landing page using Flask
Deploy the app

You might also like