Cervix ROI Segmentation Using U-NET

Last update: Sep 14, 2022

Overview

Cervix ROI Segmentation Using U-NET

Overview

This code illustrate how to segment the ROI in cervical images using U-NET.

The ROI here meant to include the: Os + transformation zone + nearby tissue.

The localized ROI is supposed to improve the classification of cervical types, which is the challenge in the Kaggle competition:Intel and MobileODT Cervical Cancer Screening

Compare to other UNET examples, in this one we got:

the input images in RGB
the input images and masks are augmented in pairs using izip ImageDataGenerators
support both Tensorflow and Theano backend, and is using Keras 2

Dependencies:

Keras 2
Tensorflow or Theano
cv2

Other references:

Thanks to Paul, who annotated the bounding box here: Bounding boxes for Type_1. I have made some minor adjustments and added the missing images.
The amazing Keras UNET implementation by jocicmarko in: ultrasound-nerve-segmentation
The original U-Net design: U-Net: Convolutional Networks for Biomedical Image Segmentation.

Usage

Data preparation:

Download the data from Kaggle.
Unzip trian.7z and test.7z into input folder. You may unzip additional_Type_*_v2.7z as well, if you want to segment them, its optional.
The input folders should look like this:
- input/test/
- input/train/Type_1
- input/train/Type_2
- input/train/Type_3
- input/additional/Type_1 (optional)
- input/additional/Type_2 (optional)
- input/additional/Type_3 (optional)
Run prepare_data.py
Run split_data.py
Note:
- The bbox annotations were converted to Sloth json format and is included under input/*.json.
- The additional data is NOT used in this training.

Training:

Run train.py
The best epoch weight file will be save under src/unet_xxxxxx/weights.h5. Note when train.py starts, it will look for previous weight file (if any) and resume from there if weight file exits

Segmentation:

Run predict.py
The output segmentations are under:
- input/test_roi/
- input/train_roi/
- input/additional_roi/

Configurations:

Customize configurations.py

Results

On a GTX 1070, the training of 400 epochs took ~2 hours to complete. The best DICE coefficient is ~0.78.

Apply this model to the 512 unseen test images, the result looks satisfactory in 96% of images.

Sample outputs:

Training loss:

You might also like...

Code for paper "ASAP-Net: Attention and Structure Aware Point Cloud Sequence Segmentation"

ASAP-Net This project implements ASAP-Net of paper ASAP-Net: Attention and Structure Aware Point Cloud Sequence Segmentation (BMVC2020). Overview We i

26 Aug 25, 2022

A PyTorch implementation for V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation

A PyTorch implementation of V-Net Vnet is a PyTorch implementation of the paper V-Net: Fully Convolutional Neural Networks for Volumetric Medical Imag

606 Dec 21, 2022

Official and maintained implementation of the paper "OSS-Net: Memory Efficient High Resolution Semantic Segmentation of 3D Medical Data" [BMVC 2021].

OSS-Net: Memory Efficient High Resolution Semantic Segmentation of 3D Medical Data Christoph Reich, Tim Prangemeier, Özdemir Cetin & Heinz Koeppl | Pr

23 Sep 21, 2022

This repository contains the code for using the H3DS dataset introduced in H3D-Net: Few-Shot High-Fidelity 3D Head Reconstruction

H3DS Dataset This repository contains the code for using the H3DS dataset introduced in H3D-Net: Few-Shot High-Fidelity 3D Head Reconstruction Access

72 Dec 10, 2022

Neural-net-from-scratch - A simple Neural Network from scratch in Python using the Pymathrix library

A Simple Neural Network from scratch A Simple Neural Network from scratch in Pyt

2 Jan 7, 2022

Mae segmentation - Reproduction of semantic segmentation using masked autoencoder (mae)

ADE20k Semantic segmentation with MAE Getting started Install the mmsegmentation

97 Dec 17, 2022

Comments

Where is the manual labels?

Thanks for your sharing. After downloading the training and testing dataset, I notice that I cannot find the manual labels. Where are they? Don't we need them?

Thanks,

opened by lowkeygit 1
Doubt in predict.py

Hi Scotty - Tks for the amazing code, helped me a lot , had a doubt in predict.py . Is it possible to explain what was done in predict.py at a high level, any references to any literature on these methods would be great Thanks -

opened by ronroc 1

Cervix ROI Segmentation Using U-NET

Related tags

Overview

Cervix ROI Segmentation Using U-NET

Overview

Usage

Results

You might also like...

Code for paper "ASAP-Net: Attention and Structure Aware Point Cloud Sequence Segmentation"

A PyTorch implementation for V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation

Official and maintained implementation of the paper "OSS-Net: Memory Efficient High Resolution Semantic Segmentation of 3D Medical Data" [BMVC 2021].

[NeurIPS2021] Code Release of K-Net: Towards Unified Image Segmentation

Code for "FPS-Net: A convolutional fusion network for large-scale LiDAR point cloud segmentation".

Implementation of U-Net and SegNet for building segmentation

This repository contains the code for using the H3DS dataset introduced in H3D-Net: Few-Shot High-Fidelity 3D Head Reconstruction

Neural-net-from-scratch - A simple Neural Network from scratch in Python using the Pymathrix library

Mae segmentation - Reproduction of semantic segmentation using masked autoencoder (mae)

Comments

Where is the manual labels?

Doubt in predict.py

Releases(v1.0)

v1.0(Jun 13, 2017)

Owner

Scotty Kwok

Deal or No Deal? End-to-End Learning for Negotiation Dialogues

Neuron Merging: Compensating for Pruned Neurons (NeurIPS 2020)

Deep generative modeling for time-stamped heterogeneous data, enabling high-fidelity models for a large variety of spatio-temporal domains.

PocketNet: Extreme Lightweight Face Recognition Network using Neural Architecture Search and Multi-Step Knowledge Distillation

GenGNN: A Generic FPGA Framework for Graph Neural Network Acceleration

Lightweight tool to perform MITM attack on local network

DualGAN-tensorflow: tensorflow implementation of DualGAN

A Unified Framework and Analysis for Structured Knowledge Grounding

Face Recognition and Emotion Detector Device

SparseML is a libraries for applying sparsification recipes to neural networks with a few lines of code, enabling faster and smaller models

This is our ARTS test set, an enriched test set to probe Aspect Robustness of ABSA.

Cancer Drug Response Prediction via a Hybrid Graph Convolutional Network

SOTA model in CIFAR10

Pytorch Implementation of LNSNet for Superpixel Segmentation

null

Implementation of Transformer in Transformer, pixel level attention paired with patch level attention for image classification, in Pytorch

Pytorch library for seismic data augmentation

Repo for CReST: A Class-Rebalancing Self-Training Framework for Imbalanced Semi-Supervised Learning

Neural Style and MSG-Net

Accelerate Neural Net Training by Progressively Freezing Layers