yashkhandelwal2006/half-unet-for-medical-image-segmentation

Half-UNet

Problem Statament

Medical image segmentation plays a vital role in computer-aided diagnosis procedures. The main goal of segmenting this data is to identify areas of the anatomy required for a particular study or medical diagnosis. U-Net is widely used in medical image segmentation. Many variants of UNet have been proposed, which attempt to improve the network performance while keeping the U-shaped structure unchanged. However, this U-shaped structure is not necessarily optimal. In this project, we attempt to analyze the segmentation ability of the existing UNet architecture and finally a more efficient architecture, Half-UNet and its variants are proposed. The proposed architectures are also encoder-decoder type architectures (similar to UNet) but in the Half-UNet architecture both the encoder block as well as the decoder block have been simplified. Furthermore we propose another architecture based on Half-UNet architecture as well as the U^2-Net Architecture which in our testing gave even better results. We compared all these models on a single left ventricular MRI dataset and all the models looked equally efficient with some having very few trainable parameters.

Implemented Half-UNet Architecture

Proposed Nested Half-UNet Architecture

Results

Evaluation

Project Structure Information

data/ - This directory contains links to data and trained weights

sources/ - This directory contains research papers implemented / used

doc/ - This directory contains implementation details for this project

src/ - This directory contains code files for data pre-processing, model training and model evaluation for all the above mentioned models

half-unet-for-medical-image-segmentation's Issues

How to download the dataset?

Thank you for your nice code. Allow me to ask more after I find it but have not seen it yet.
Plz help me how to download the data with the label?
1/ I followed the reference [9]

2/ and had like this but not have label to training.

Thanks.