Creation and evaluation of Convolutional neural networks to segment Brain Tumours in MRI scans.
Using a U-Net architecture and ResNet encoder pre trained on ImageNet.
It is trained on 3,064 patients records and uses PyTorch.
-
Install uv package manager.
-
Install required dependencies
uv syncAlternatively requirements.txt has also been generated if uv cannot be used.
Evaluation of the models are available here in notebook format.
Evaluation of the preprocessing is available here in notebook format.
This preprocesses, loads and trains the model based on the provided configuration.
It uses whatever accelerator is available - GPU, CPU.
The weights can be saved to a file for evaluation or further processing.
uv run pipeline.py With optional arguments documented in pipeline.py:
uv run pipeline -subsetsize=1000 -batchsize=32 -epochs=50 -modelpath=resnet18.pth