wdbm / dosimetric_correlations

dosimetric_correlations

note

This project is a work in progress.

setup

Install the following:

• NumPy
• pandas
• Scikit-learn
• TensorBoard
• TensorFlow

use

The input data could be CSV with the following fields:

• features:
  • PTV vol (cc)
  • Lungs-GTV vol (cc)
  • Lungs-GTV-PTV vol (cc)
  • Lungs-GTV in PTV vol (cc)
  • KBP Lungs (cc)
  • Lungs-GTV - KBP Lungs (cc)
  • Heart vol (cc)
  • Heart in PTV vol (cc)
• targets:
  • V5 (%)
  • V20 (%)
  • Mean Lungs-GTV (Gy)
  • V30 (%)
  • Mean (Gy)

It could be CSV with the following fields:

• features:
  • Dose/#
  • Prescription
  • PTV vol (cc)
  • Lungs-GTV vol (cc)
  • Lungs-GTV-PTV vol (cc)
  • Lungs-GTV in PTV vol (cc)
  • KBP Lungs (cc)
  • Lungs-GTV - KBP Lungs (cc)
• targets:
  • V5 (%)
  • V20 (%)
  • Mean Lungs-GTV (Gy)

The rightmost columns should be the targets. The number of targets can be specified as an argument for a neural network script.

Manually remove missing values from data.

Preprocess the CSV data such that all features are scaled to (-1, 1):

./preprocess_CSV_file.py --infile=data.csv --outfile=preprocessed_data.csv

Train and evaluate on preprocessed CSV data with TensorBoard:

./cures_cancer.py --help

./cures_cancer.py --infile=preprocessed_data.csv --TensorBoard