Prostate segmentation task on IDC collections
The goal of this project is to augment IDC collections with DL based segmentation masks obtain from nnU-net framework, using pre-trained models inference on IDC collections.
Two nnU-net based pre-trained models have been used for inference on two different IDC collections.
The pre-trained models, Task005-Prostate(PZ and TZ) and Task024-Promise(whole prostate) can be found/downloaded in Zenodo.
The two labelled IDC collections are ProstateX and QIN-PROSTATE-Repeatability.
The nnU-net architecture model chosen is 3d_fullres across all experiments, with test-time data augmentation enabled. Task024 has only one input modality, T2, whereas Task005 is multi-modal, T2 and ADC.
All these ipynb were run through google colab.
Below is detailed description of the structure of the repository :
-
- idc-prostate_segmentation_promised24_QIN-PROSTATE-Repeatability.ipynb : Experiment on QIN-PROSTATE-Repeatability using pre-trained model Task024-Promise.
- idc-prostate_segmentation_promised24_prostateX.ipynb : Experiment on ProstateX using pre-trained model Task024-Promise.
- idc-prostate_segmentation_prostate05_qin-rep-repeat.ipynb : Experiment on Qin-Prostate-Repeatability using pre-trained model Task005-Prostate computing evaluation metrics on Peripheral Zone of the prostate only
- results_analysis.ipynb : Results analysis from experiments
-
nnunet_results : Contains the DSC and other quantitative metrics results for each experiment, used also in results_analysis.ipynb
-
Open any of these ipynb in Google Colab!