Skip to content

Latest commit

 

History

History
130 lines (90 loc) · 7.25 KB

README.md

File metadata and controls

130 lines (90 loc) · 7.25 KB

Attention 3D UnetPP Brain-Tumor-Segementation Ensemble Learning

this repo contain the implementation of paper we are working with the team from Jorden Univesity of science and technology , in this work we present a new based model on Neasted 3DUnet++ combined with attention mechanism Block for more effiency feature extracttion

the problem case study in 3D Biomedical image processing

  • subject

Accurate segmentation of brain tumor sub-regions is essential in the quantification of lesion burden, providing insight into the functional outcome of patients. In this regard, 3D multi-parametric magnetic resonance imaging (3D mpMRI) is widely used for non-invasive visualization and analysis of brain tumors. Different MRI sequences (such as T1, T1ce, T2, and FLAIR) are often used to provide complementary information about different brain tumor sub-regions

  • Problem

in many cases for processing Medical images to get better understanding of disease and impact on human being life such Brain tumor is most area for reseachers to improve system diagnosis in partuclar Task Segementation , last few years lunch of challenges BRATS for segmentation Brain tumor Sub-regrion many of studying came up to improve CAD system ,

  • Solution

For automatic segmentation we will use Unet3d To predict the 3D Volumitric Shape of tumor: we used Neasted 3DUnet++ combined with attention mechanism Block for more effiency feature extracttion and we trained on indenpendently 3DUnet++ and 3DUnet++ combined with attention mechanism

  1. introduction
  2. environment project
  3. run project
  4. model
  5. Results
  6. Prediction

introduction

Imaging Data Description

All BraTS multimodal scans are available as NIfTI files (.nii.gz) and describe a) native (T1) and b) post-contrast T1-weighted (T1Gd), c) T2-weighted (T2), and d) T2 Fluid Attenuated Inversion Recovery (T2-FLAIR) volumes, and were acquired with different clinical protocols and various scanners from multiple (n=19) institutions, mentioned as data contributors here.

All the imaging datasets have been segmented manually, by one to four raters, following the same annotation protocol, and their annotations were approved by experienced neuro-radiologists. Annotations comprise the GD-enhancing tumor (ET — label 4), the peritumoral edema (ED — label 2), and the necrotic and non-enhancing tumor core (NCR/NET — label 1), as described both in the BraTS 2012-2013 TMI paper and in the latest BraTS summarizing paper. The provided data are distributed after their pre-processing, i.e., co-registered to the same anatomical template, interpolated to the same resolution (1 mm^3) and skull-stripped.

environment-project

  • setup the enviroment;

    • install script shell here you will need to run script shell to install all the dependencies needed for run code :

      • create kaggle account to access to the data API

      • add path kaggle.json to script shell $path_api

      • create the enviromenet here you will need to run

           conda create --name Segemnetation python=3.6

      • make sure the requirements.txt exist to the repo install the packges if you want fisrt neeed to run

           pip install -r requirements.txt

      • here you will need to run script shell to install all the dependencies needed automated setup whole project

           chmod +x download_dataset.sh && ./automate_downlaod_data.sh

run-project

  • PostProcessig dataset Brast2020;

    this tool built based on top of BET algorithm that publish from FSL and N4baisCorrection we automated the process and handle the data in 3D shape link Project Repository visited and run this First Stage

Model-Description

our Pipline divde into tow Stage to train model Unet++ and Unet++ with Attention Gate

  • The Full Pipline Ensemble learning:

    in our Purposal we trained the model indenpendently and we combined the prediction both of the wieghts Checkpoints we provide open to downlaod following line Drive

    Models Parameters Size MB
    3D Attention Unet++ 30 M 101.5 MB
    3D Unet++ 31 M 785 MB

    Stratgy used is Wieght Voting algorithm here's the full picture

  • Ruuning the model indenpendently:

    we provide the Yaml file configuration for Fine-Tuning the model brast2020.yml , has all the Hyper-Parameters to setup , we can change the model by updating the Model in config file which are [UNET3DPPATTEN ,UNET3DPP ] for run the model to train on GPU follwing the command :

        python train.py --stage train --config brast2020.yaml --fold 0

  • Evaluation the model indenpendently:

    we provide here also the evaluation model indenpendently , we can change the model by updating the Model in config file evaluate.yaml which are [UNET3DPPATTEN ,UNET3DPP ] for run the model to train on GPU follwing the command :

        python predict.py --stage evaluate --config evaluate.yaml

Results

the performence of the both model we trained on indenpendently for 250 Epochs shows in the table and Diagram blow :

Table 1: show the Evaluation Metrics we used DCS|JSC

Models Dice Similarity WT Dice Similarity TC Dice Similarity ET Jaccard Similarity WT Jaccard Similarity TC Jaccard Similarity ET
Pipline A 0.88 0.87 0.73 0.79 0.78 0.59
Pipline B 0.82 0.82 0.69 0.72 0.72 0.54
Ensemble Learning 0.86 0.86 0.71 0.77 0.77 0.57

Figure : the virtualization of Results Bar Plot both of UNET3DPPATTEN ,UNET3DPP


UNET3DPPATTEN ,UNET3DPP comparisiont

Figure : Results of Ensemble Wieght Voting Bar Plot both of Ensemble Learning


Ensemble Weight Voting Bar Plot Ensemble Wieght

Prediction

we have a full Notebook has all the Quantative results used both modesl we trained on with Ensemble Learning results-attention-3d-unetpp-torchio

here the 3D prediction of Enesmeble Learning model both of Animation Slicer Frames and 3D Extraction Volumitric Data

  • Animations Frames:
  • Animations 3D Volumitric Data: