Localization of clinically significant prostate cancer on multiparametric MR: an open reproducible analysis on the digitalized PROMIS dataset

Welcome to the repository for the fully digitalised open-source PROMIS study, along with the preprocessing tools and code for reproducing the quantitative analysis for two studies:

• A patient-level diagnostic accuracy described in the original paper [Ahmed et al 2017], Diagnostic accuracy of multi-parametric MRI and TRUS biopsy in prostate cancer (PROMIS): a paired validating confirmatory study, and
• A follow-up analysis on quantifying zone-level and lesion-level localization accuracy: [anonymous] (under review; will be updated upon publication), Localization of clinically significant prostate cancer on multiparametric MR: an open reproducible analysis on the digitalized PROMIS dataset.

What's included

• PROMIS dataset: An open-source fully digitalised dataset curated from the PROMIS study, including aligned radiological and histopathological labels.
• Preprocessing tools: Tools to prepare the dataset for subsequent automated quantitative localisation analysis and potentially machine learning tasks.
• Diagnostic accuracy analysis: Code to reproduce the main results reported in the above two studies with their statistical analysis.

The PROMIS dataset

Overview

Item	Description
Image Data Modality	T2-weighted, High-b DWI, ADC
Image Data Format	NifTi
Image Annotations	Lesion countours, prostate gland mask
Clinical report	Template biopsy report, radiologist readings

Download

You can download the dataset here.

Generating local zones on the prostate masks

To generate local zones of different granularity, including hemi, quadrant, octant and Bazell zones, run the following script:

python gen_localised_zones.py

Diagnostic accuracy at patient-level and zone-levels

To compute the main analysis results:

1. Specify configuration
   Define all required variables and directory paths in the config.py file. This includes paths to the dataset, output directories, and any relevant parameters.

2. Run analysis
   Execute the main analysis script:

   python localised_analysis.py