Vessel-Involvement-Quantifier

Pancreatic ductal adenocarcinoma (PDAC) resectability assessment often suffers from inconsistent manual evaluations of vascular involvement in CT scans. This study introduces VasQNet, an AI model trained on a dataset of 613 CTs from PDAC patients and controls. VasQNet autonomously segments PDAC and vasculature, quantifies vascular involvement, and classifies resectability stages (resectable, borderline, or locally advanced) based on Dutch Pancreatic Cancer Group criteria. In testing, VasQNet demonstrated strong agreement with expert radiologist assessments, showing promise for enhancing the reliability and objectivity of PDAC resectability determinations in clinical settings. VasQNet makes the following contributions to the field:

1. Automated Vascular Assessment: VasQNet automates the quantification of vascular involvement in PDAC using AI-segmented CT scans, reducing subjectivity.
2. Resectability Classification: VasQNet classifies PDAC into resectable, borderline resectable, or locally advanced stages, aiding treatment decisions.
3. Improved Consistency: By reducing interobserver variability, VasQNet enhances agreement among radiologists, leading to more reliable assessments.

Installation and Setup

The VesselInvolvementQuantifier has been tested on MacOS (Monterey, Version 12.6) and Windows 11. We do not provide support for other operating systems.

The VesselInvolvementQuantifier does not require a GPU.

We very strongly recommend you install the VesselInvolvementQuantifier in a virtual environment.

Python 2 is deprecated and not supported. Please make sure you are using Python 3.

For more information about the VesselInvolvementQuantifier, please read the following paper:

TODO: add citation here

Please also cite this paper if you are using the VesselInvolvementQuantifier for your research!

Follow these steps to run the VesselInvolvementQuantifier:

1. Install the VesselInvolvementQuantifier

 git clone https://github.com/PHAIR-Amsterdam/Vessel-Involvement-Quantifier.git
 cd Vessel-Involvement-Quantifier
 pip install -e .

2. The VesselInvolvementQuantifier needs to know which vessel segments to analyze and which degree and resectability categories to use. For this you need to set a variables in settings.py :

   2.1 Save the key file (xlsx) in the Vessel-Involvement-Quantifier folder. The key file has the name db_vessels.xlsx and contains the patient identifiers (column numbber) and the ground truth involement classifcations for each vessel segment you intend to analayze.

   2.2 Save the segmentation masks you intend to analyze in the Vessel-Involvement-Quantifier/test_data folder. The filenames must contain the patient identifier used in your key file. The file must be of type .nii.gz or *.nrrd.

   2.3 Set the vessel segment names as used in the column headers of your key file in settings.py under vessels. Set the segment index in you segmentation mask under vessels_id.

   2.4 Set the degree classifcations provided by your radiologist under in settings.py under degrees_id.

   2.5 Set the resectability classifcations provided by your radiologist under in settings.py under resectability_id.

VesselInvolvementQuantifier
├── test_data
│   └── patient1.nrrd
│   └── patient2.nrrd
├── db_vessels.xlsx
├── main.py
├── settings.py
├── plot.py
├── quantify.py
├── utils.py

Analyzing

After following the instruction under Installation (2) to store your test data and key file, run the function main.py. The VesselInvolvementQuantifier will store results in results.csv and will create a boxplot under the name boxplot.png.