/ISeeU

ISeeU: Visually interpretable deep learning for mortality prediction inside the ICU

Primary LanguageJupyter NotebookMIT LicenseMIT

ISeeU: Visually interpretable deep learning for mortality prediction inside the ICU

Note: This version of ISeeU has been tested with numpy 1.12.1, pandas 0.23.4, keras 2.2.4, deeplift 0.6.6.2, matplotlib 2.0.2 and tensorflow 1.9.0.

A ConvNet trained on MIMIC-III data for mortality prediction inside the Intensive Care Unit. It uses a set of 22 predictors sampled during the first 48h of ICU stay to predict the probability of mortality. This set of predictors roughly corresponds to those used by the SAPS-II severity score:

  • AGE
  • AIDS
  • BICARBONATE
  • BILIRRUBIN
  • BUN
  • DIASTOLIC BP
  • ELECTIVE
  • FiO2
  • GCSEyes
  • GCSMotor
  • GCSVerbal
  • HEART RATE
  • LYMPHOMA
  • METASTATIC CANCER
  • PO2
  • POTASSIUM
  • SODIUM
  • SURGICAL
  • SYSTOLIC BP
  • TEMPERATURE
  • URINE OUTPUT
  • WBC

ISeeU achieves 0.8735 AUROC when evaluated on MIMIC-III. More information is available in our paper. It also can be installed from PyPi:

pip install iseeu