INTEGRAL Radiomics

This repository contains the model and example code for using the INTEGRAL-Radiomics screen-detected pulmonary nodule malignancy model reported in:

Warkentin MT, Al-Sawaihey H, Lam S, et al Radiomics analysis to predict pulmonary nodule malignancy using machine learning approaches Thorax Published Online First: 09 January 2024. doi: 10.1136/thorax-2023-220226

If you have any comments or questions, please file an Issue.

Usage

Setup

To use this model, we assume you have performed feature extraction using the PyRadiomics Python library (https://pyradiomics.readthedocs.io). Note, that this study used PyRadiomics V.3.0.1. More information about the feature extraction can be found in the published manuscript cited below.

We provie the YAML configuration file (PyRadiomics_config.yaml) in this repository so that an identical feature extraction can be performed. See the PyRadiomics website for details on performing customized feature extractions (https://pyradiomics.readthedocs.io).

In addition to the radiomics features, the following nine patient-level features are required:

• epi_age: Patients age (years)
• epi_female: Binary variable for sex (0=Male, 1=Female)
• epi_fhlc: Binary variable for family history of lung cancer (0=No, 1=Yes)
• epi_copdemph: Binary variable for history of COPD or emphysema (0=No, 1=Yes)
• epi_formersmk: Former smoker status (0=No, 1=Yes)
• epi_duration Number of years smoked cigarettes
• epi_cigday: Cigarettes per day
• epi_quittime: Years since quitting for former smokers (Set to 0 for current smokers)
• epi_bmi: Body mass index (kg/m^2)

The data frame MUST also contain identifying variables for study (study ID), pid (patient ID), and nid (nodule ID), though these may be set to missing (NA) or use random values as these data are not used in the prediction but must be in the data frame for preprocessing.

Note: The data should NOT be normalized/standardized/scaled prior to using the predict() function described below. Normalization of predictors happens automatically during the call to predict().

Model Predictions

Once the feature extraction is complete, the code presented below can be used to load the model and make predictions on your data frame (or tibble), which is named your_data in the example below.

As described in the previous section, the data frame for prediction must contain the three ID variables, nine patient variables, and the radiomics features. All other columns will be ignored.

# install.packages(c('parsnip', 'recipes', 'workflows', 'glmnet', 'vetiver'))
library(parsnip)
library(recipes)
library(workflows)
library(glmnet)
library(vetiver)

# Load the INTEGRAL-Radiomics model
integral_rad <- readRDS('INTEGRAL-Radiomics.rds')

# Predict probabilities for `your_data`
predict(integral_rad, new_data = your_data, type = 'prob')

The predict(...) function will return a data frame with two columns (.pred_0 and .pred_1) that correspond to the probabilties of a pulmonary nodule being benign (.pred_0) or malignant (.pred_1). Users may wish to threshold these probabilities to obtain binary class labels.

Citation

If you use this model, please cite the following article:

Warkentin MT, Al-Sawaihey H, Lam S, et al Radiomics analysis to predict pulmonary nodule malignancy using machine learning approaches Thorax Published Online First: 09 January 2024. doi: 10.1136/thorax-2023-220226

@article {Warkentinthorax-2023-220226,
  author = {Matthew T Warkentin and Hamad Al-Sawaihey and Stephen Lam and Geoffrey Liu and Brenda Diergaarde and Jian-Min Yuan and David O Wilson and Sukhinder Atkar-Khattra and Benjamin Grant and Yonathan Brhane and Elham Khodayari-Moez and Kiera R Murison and Martin C Tammemagi and Kieran R Campbell and Rayjean J Hung},
  title = {Radiomics analysis to predict pulmonary nodule malignancy using machine learning approaches},
  elocation-id = {thorax-2023-220226},
  year = {2024},
  doi = {10.1136/thorax-2023-220226},
  publisher = {BMJ Publishing Group Ltd},
  issn = {0040-6376},
  URL = {https://thorax.bmj.com/content/early/2024/01/08/thorax-2023-220226},
  eprint = {https://thorax.bmj.com/content/early/2024/01/08/thorax-2023-220226.full.pdf},
  journal = {Thorax}
}