|
|
--- |
|
|
license: mit |
|
|
language: |
|
|
- en |
|
|
metrics: |
|
|
- accuracy |
|
|
- precision |
|
|
pipeline_tag: image-classification |
|
|
tags: |
|
|
- biology |
|
|
- cancer |
|
|
- glioblastoma |
|
|
- brain |
|
|
- multimodal |
|
|
- radiogenomics |
|
|
- radiomics |
|
|
- immune |
|
|
- classifier |
|
|
--- |
|
|
<p align="center"> <b> Predictive Radiomics for Evaluation of Cancer Immune SignaturE in Glioblastoma | PRECISE-GBM </b> </p> |
|
|
|
|
|
<p align="center"> |
|
|
<img src="PRECISE-GBM_GUI_logo%20(1).png" alt="PRECISE-GBM Logo"> |
|
|
</p> |
|
|
|
|
|
[](https://opensource.org/licenses/MIT) |
|
|
|
|
|
This repository contains an AI-based training and retraining pipeline for Predictive Radiomics for Evaluation of Cancer Immune SignaturE in Glioblastoma (PRECISE-GBM). It is the multimodal radiogenomic framework that integrates MRI radiomics, genomics, and immune signatures for patient stratification. |
|
|
|
|
|
<b> Project: PRECISE-GBM - Model training & retraining helpers </b> |
|
|
|
|
|
## 📝 Overview |
|
|
|
|
|
This repository contains code to train models (Gaussian Mixture labelling + SVM and ensemble classifiers) and to persist all artifacts required to reproduce or retrain models on new data. It includes: |
|
|
|
|
|
- `Scenario_heldout_final_PRECISE.py` — training pipeline producing `.joblib` models and metadata JSONs (selected features, best params, CV results). |
|
|
- `retrain_helper.py` — CLI utility to rebuild pipelines, set best params and retrain using saved selected-features and params JSONs. Supports JSON/YAML config files and auto-detection of model type. |
|
|
- `README_RETRAIN.md` — detailed retrain examples and a notebook cell. |
|
|
|
|
|
This repo also includes helper files to make it ready for GitHub: |
|
|
- `requirements.txt` — Python dependencies |
|
|
- `.gitignore` — recommended ignores (models, caches, logs) |
|
|
- `LICENSE` — MIT license |
|
|
- GitHub Actions workflow for CI (pytest smoke test) |
|
|
|
|
|
## 📝 Getting started (Windows PowerShell) |
|
|
|
|
|
1) Create and activate a virtual environment |
|
|
|
|
|
```powershell |
|
|
python -m venv .venv |
|
|
.\.venv\Scripts\Activate.ps1 |
|
|
``` |
|
|
|
|
|
2) Install dependencies |
|
|
|
|
|
```powershell |
|
|
pip install --upgrade pip |
|
|
pip install -r requirements.txt |
|
|
``` |
|
|
|
|
|
3) Run training (note: the training script reads data from absolute paths configured in the script — adjust them or run from an environment where those files are present) |
|
|
|
|
|
```powershell |
|
|
python Scenario_heldout_final_PRECISE.py |
|
|
``` |
|
|
|
|
|
The training script will create model files under `models_LM22/` and `models_GBM/` and write metadata JSONs next to each joblib model (selected features, params, cv results) as well as group-level JSON summaries. |
|
|
|
|
|
## 📝 Retraining |
|
|
|
|
|
See `README_RETRAIN.md` for detailed CLI and notebook examples. Short example: |
|
|
|
|
|
```powershell |
|
|
python retrain_helper.py \ |
|
|
--model-prefix "models_GBM/scenario_1/GBM_scen1_Tcell" \ |
|
|
--train-csv "data\new_train.csv" \ |
|
|
--label-col "label" |
|
|
``` |
|
|
|
|
|
## 📝Notes |
|
|
|
|
|
- The training script contains hard-coded absolute paths to data files. Before running on another machine, update the `scenarios_*` file paths or place the datasets in the same paths. |
|
|
- Retrain helper auto-detects model type when `--model-type` is omitted by looking for `{prefix}_svm_params.json` or `{prefix}_ens_params.json`. |
|
|
- YAML config support for retrain requires PyYAML (`pip install pyyaml`). |
|
|
|
|
|
## 📝 CI |
|
|
|
|
|
A basic GitHub Actions workflow runs a smoke pytest to ensure the retrain helper imports and basic pipeline construction works. It does not run heavy training. |
|
|
|
|
|
## 📝 Contributing |
|
|
|
|
|
See `CONTRIBUTING.md` for guidance on opening issues and PRs. |
|
|
|
|
|
## 📝 License |
|
|
|
|
|
This project is released under the MIT License — see `LICENSE`. [MIT License](https://opensource.org/licenses/MIT). |
|
|
|
|
|
## 📝 Citation |
|
|
Please use the following citations when using the repository. |
|
|
|
|
|
2025 |
|
|
|
|
|
> **Ghimire P, Modat M, Booth T**. *Predictive radiogenomic AI Model for patient stratification in brain tumor immunotherapy trials. Neuro-oncology. Oct 2025; 26(Suppl_3): iii58–iii59. doi: https://doi.org/10.1093/neuonc/noaf193.188* |
|
|
|
|
|
> **Ghimire P, Modat M, Booth T**. *Radiogenomic AI model predicts immune status in IDH wildtype glioblastoma: PRECISE-GBM study. RCR open. Jan 2025; 3(1): 100234* |
|
|
|
|
|
2024 |
|
|
|
|
|
> **Ghimire P, Modat M, Booth T**. *A machine Learning bases predictive radiomics for evaluation of cancer immune signature in glioblastoma: the PRECISE-GBM study. Neuro-Oncology. Oct 2024; 26(suppl_5): v25.* |
|
|
|
|
|
> **Ghimire P, Modat M, Booth T**. *A radiogenomic machine learning based study to identify Predictive Radiomics for Evaluation of Cancer Immune SignaturE in IDHw Glioblastoma. Neuro-Oncology. Oct 2024; 26(suppl_7): vii3* |
|
|
|
|
|
|
|
|
**Contact**: |
|
|
|
|
|
**Dr Prajwal Ghimire** |
|
|
|
|
|
**MBBS MRCSEd MSc PhD'26** |
|
|
|
|
|
School of Biomedical Engineering & Imaging Sciences, King's College London |
|
|
|
|
|
Email: [prajwal.1.ghimire@kcl.ac.uk](mailto:prajwal.1.ghimire@kcl.ac.uk) |
