Hippocampus-Segmentation-and-Classification-in-Alzheimer-s-disease-using-Deep-CNN

This file includes the description of the project and instruction to run the codes.

Objective: To build a deep learning framework with CNN models that performs hippocampal segmentation and disease classification in Alzheimer’s disease

Alzheimer’s disease (AD) is a progressive and irreversible brain degenerative disorder characterized by memory loss and cognitive impairment. At present, there are more than 90 million patients diagnosed with AD. By 2050, it is estimated that the number of AD patients will reach 300 million. No effective treatments are available to cure AD at present, while current AD treatments and medicines can only ease symptoms of AD or slow down AD progression. Therefore, the diagnoses of AD at early stage is an effective way to prevent AD. The current clinical protocol to detect volumetric changes in the hippocampus is manual segmentation, which is time-consuming, observer-dependent and challenging. As A Result, an automated approach to hippocampus segmentation is imperative to improve the efficiency and accuracy of the diagnostic workflow. Several automatic and semi-automatic segmentation approaches have been proposed, which utilize T1-weighted structural MRIs, to segment the hippocampus.

Recently, deep learning networks, including convolutional neural networks (CNNs), have been widely used in image classification and computer vision. The deep 3D CNNs are used to extract the features of 3D medical images for classification. Deep learning has been increasingly applied in the field of biomedical image-based diagnosis. U-net is a neural network is widely used for biomedical image segmentation. It was first designed and applied in 2015 to process biomedical images. As a general CNN focuses its task on image classification, where input is an image and output is one label, but in biomedical cases, it requires us not only to distinguish whether there is a disease, but also to localize the area of abnormality. U-Net is dedicated to solving this problem. The reason it can localize and distinguish borders is by doing classification on every pixel, so the input and output share the same size.

Our network is based on the 3D U-Net, with convolutions in the lowest layer between the encoder and decoder paths, residual connections between the convolution blocks in the encoder path, and residual connections between the convolution blocks and the final layer of the decoder path. The aim of this project is to evaluate the performance of the network using the hippocampus dataset provided as part of the Medical Segmentation Decathlon and to evaluate the performance of the U-Net model by Dice similarity coefficient (DSC), Jaccard index (JI) and sensitivity metrics.

Data Acquisition:

Data is acquired from http://medicaldecathlon.com/. This dataset is stored as a collection of NIFTI files, with one file per volume, and one file per corresponding segmentation mask. The original images are T2 MRI scans of the full brain. In this dataset the cropped volumes are used where only the region around the hippocampus has been cut out. As the data provided was already truncated to the region of interest around the hippocampus, not much data pre-processing was required. There are total 260 images and 260 labels used. Each image is represented by a single file and has a corresponding label file which is named the same as the image file.

Folders and files

scr scr/result: json file that has the results of dice, JI and sensitivity

scr/run: logs files of the Tensorboard (explained in the last section)

scr/images : contains 2 nifti files (actual dataset has 260 files, needs to be downloaded from the link above before running the code)

scr/lables : contains 2 nifti files (actual dataset has 260 files, needs to be downloaded from the link above before running the code)

scr/02_code.py and 03_scr/code.ipynb

scr/01_hippo_EDA.ipynb : Data Exploratory code which includes the data explorations part of the project

result_screenshots : screenshots of the results from tensorboard and anaconda prompt

The Programming Environment:

You would need a Python 3.7+ environment with the following libraries:

numpy

nibabel

matplotlib

PIL

pydicom

json

torch (preferably with CUDA)

tensorboard

Description of the code

There are 3 codes :

01_hippo_EDA.ipynb : Data Exploratory code which includes the data explorations part of the project
02_code.py
03_code.ipynb

ipynb file is used to see and understand the code as ipynb is easy to read than the py file. So .py file is the file that is used to get the results. Both the files contain same lines of code.

The code.py file contains the below functions :

Loading and preprocessing functions:

mpl_image_grid

log_to_tensorboard

save_numpy_as_image

med_reshape

LoadHippocampusData

SlicesDataset

Evaluation methods functions

Dice3d

Jaccard3d

sensitivity

U-net functions

UNet

UnetSkipConnectionBlock

UNetInferenceAgent

single_volume_inference_unpadded

single_volume_inference

UNetExperiment

Training and testing functions

train

validate

save_model_parameters

load_model_parameters

run_test

run

Main function

Contains code that will kick off training and testing processes

How to run the code:

This code can be run in the Anaconda prompt/ Windows cmd

1.Install all the necessary libraries mentioned above

2.Clone this repository and unzip it

3.In your Anaconda prompt, go to the downloaded directory

4.Run the 02_code.py file. The model starts running displaying the epoches and loss values. It take some time to complete.

5.Once the scrolling stops, run the command : tensorboard --logdir runs --bind_all

After the command, Tensorboard writes logs into a folder called runs and you will be able to view results by opening the browser and navigating to default port 6006 of the machine where you are running it.

See the results_screenshot folder to see the screenshots of the results.

Archanam5282/Hippocampus-Segmentation-and-Classification-in-Alzheimer-s-disease-using-Deep-CNN