Fully automatic PDAC detection on CECT - PANORAMA challenge baseline


Logo for Fully automatic PDAC detection on CECT - PANORAMA challenge baseline

About

Creator:
Image Version:
bcce8bad-361b-42e0-83bc-aa8647893c1d
Last updated:
May 26, 2024, 11:26 a.m.
Inputs:
  • Venous phase CT scan  (A contrast-enhanced CT scan in the venous phase and axial reconstruction)
Outputs:
  • PDAC Likelihood  (This indicates the likelihood of pancreatic ductal adenocarcinoma (PDAC) occurring, presented as a value between 0.0 and 1.0.)
  • PDAC Detection Map  (Single-class, detection map of pancreatic ductal adenocarcinoma (PDAC) in 3D, where each voxel represents a floating point in range [0,1])

Challenge Performance

Date Challenge Phase Rank
May 28, 2024 PANORAMA Development Phase - Tuning 4
May 28, 2024 PANORAMA Development Phase - Sanity Check 1

Model Facts

Summary

This is the baseline algorithm for the PANORAMA challenge. The algorithm is based on the nnU-Net framework (v2) and consists of a two-step approach for pancreatic ductal adenocarcinoma (PDAC) detection on contrast-enhanced CT scans (CECT). First, a low-resolution nnU-Net model with Dice loss is trained to segment the pancreas. Based on this automatic segmentation, a region of interest (ROI) around the pancreas region is cropped from the original input CECT. The cropped ROIs are used to train a high-resolution nnU-Net algorithm for PDAC detection using cross-entropy loss. This process is summarized in the figure below.

For more details and to access the source code refer to this Github repository.

Mechanism

This algorithm produces a likelihood score for the presence of pancreatic ductal adenocarcinoma (PDAC) in an input venous-phase contrast-enhanced computed tomography scan (CECT), together with a detection map with the lesion localization. The algorithm was trained with 2238 images from the PANORAMA public development data set.

Validation and Performance

Uses and Directions

This algorithm was developed for research purposes only. This algorithm is intended to be used only on venous-phase CECT examinations . This algorithm should not be used in different patient demographics.

Benefits: Automatic detection and localization of PDAC with additional information regarding surrounding anatomy.

Target Population: This algorithm was trained on a cohort of 2238 patients from Radboud University Medical centre and University Medical Center Groningen, of which 676 had PDAC.

General use: This model is intended to be used by radiologists for predicting PDAC in venous-phase CECT scans. The model is not a diagnostic for cancer and is not meant to guide or drive clinical care. This model is intended to complement other pieces of patient information in order to determine the appropriate follow-up recommendation.

Appropriate decision support: The model identifies lesion X as at a high risk of being malignant. The referring radiologist reviews the prediction along with other clinical information and decides the appropriate follow-up recommendation for the patient.

Before using this model: Test the model retrospectively and prospectively on a diagnostic cohort that reflects the target population that the model will be used upon to confirm the validity of the model within a local setting.

Safety and efficacy evaluation: To be determined in a clinical validation study.

Warnings

Risks: Even if used appropriately, clinicians using this model can misdiagnose cancer. Delays in cancer diagnosis can lead to metastasis and mortality. Patients who are incorrectly treated for cancer can be exposed to risks associated with unnecessary interventions and treatment costs related to follow-ups.

Clinical rationale: The model is not interpretable and does not provide a rationale for therisk scores, beyond the localization structures presented in its output heatmap. Clinicians are expected to place the model output in context with other clinical information to make the final determination of diagnosis.

Discontinue use if: Clinical staff raise concerns about the utility of the model for the intended use-case or large, systematic changes occur at the data level that necessitates re-training of the model.

Common Error Messages

Information on this algorithm has been provided by the Algorithm Editors, following the Model Facts labels guidelines from Sendak, M.P., Gao, M., Brajer, N. et al. Presenting machine learning model information to clinical end users with model facts labels. npj Digit. Med. 3, 41 (2020). 10.1038/s41746-020-0253-3