Abdominal CT based Axial 2D Kidney Segmentation Pytorch : Implementation of U-Net

Abdominal CT Axial based left kidney, right kidney segmentaion Unet model.

This is my first computer vision deep learning project.

Please let me know if you find my mistake!

Also, if you want to advise me, please contact me.

contact : [email protected]

Dataset (KiTS 23)

https://kits-challenge.org/kits23/

Installation

git clone https://github.com/neheller/kits23
cd kits23
pip3 install -e .

Data Download

kits23_download_data

Original Class : background(0), kidney(1), tumor(2), cyst(3) changed to background(0), left kidney(1), right kidney(2)

Model Outputs

Metrics

Dice Loss

IoU(Intersection over Union)

Experiments Table

Exp 1. optimizer = Adam / lr = 0.0001 / loss function = Dice Loss / Batch Size =16 / Epoch = 100

Exp 2. optimizer = Adam / lr = 0.0001 / loss function = Dice CE Loss / Batch Size =16 / Epoch = 100

Exp 3. optimizer = AdamW / lr = 0.0001 / loss function = Dice Loss / Batch Size =16 / Epoch = 100

Exp 4. optimizer = AdamW / lr = 0.0001 / loss function = Dice CE Loss / Batch Size =16 / Epoch = 100

All Experiments ignore background in calculating loss.