Authors: Katy Scott, Caryn Geady, Kaitlyn Kobayashi
Contact: [email protected]
Description: Simulation code accompanying the Retiring the Ruler: Toward a Volumetric, Multi-Lesion, Longitudinal Lens on Tumor Response paper.
Pixi is required to run this project. If you haven't installed it yet, follow these instructions
- Clone this repository to your local machine
- Navigate to the project directory
- Set up the environment using Pixi:
pixi install- Artificially generate a dataset of N patients (input arg: N).
- For each patient, generate a random number of lesions (range: 1-30, poisson distribution with µ = 5).
- For each lesion, generate a random diameter (select from a list of possible diameters, with replacement (using original_shape_Maximum2DDiameterSlice)).
- For each lesion, generate a diameter change (select from a Gaussian distribution, with fractional changes from -1 to 1).
- For each lesion, assign a location tag (e.g. liver, lung, bone, brain, again from a list of possible locations, with replacement).
- For each patient, select a random number of lesions to be target lesions (up to 5 per patient, with a maximum of 2 per assigned location).
- From the selected target lesions, assess the RECIST response (SLD = 100% decrease = CR, SLD > 30% decrease = PR, SLD > 20% increase = PD, otherwise SD).
- For each patient, assess the RECIST response across all lesions (as above).
- Compare the RECIST response across all lesions to the RECIST response across target lesions only.
- For each diameter, calculate the corresponding volume (assuming spherical shape; this will represent the lower end of volumetric measurement).
- Inject inter-observer variability by adding a random error to the volume calculation (select from a Gaussian distribution, with fractional diameter changes from -0.113 to 0.113; REF: https://pmc.ncbi.nlm.nih.gov/articles/PMC3423763/).
- For each lesion, calculate the volume using another diameter measurement (i.e., original_shape_Maximum3DDiameter).
Click here to view the full documentation.