Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Global Radiomics to Parametric Maps: A Unified Workflow Fusing Radiomics and Deep Learning for PDAC Detection

Zengtian Deng, Yimeng He, Yu Shi, Lixia Wang, Touseef Ahmad Qureshi, Xiuzhen Huang, Debiao Li

TL;DR

This work addresses PDAC detection by uniting handcrafted radiomics with deep learning through a radiomics-aware, two-stage nnUNet. The pipeline first selects a compact set of discriminative global radiomic features from the whole pancreas and then fuses these features, along with voxel-wise radiomic maps, into a two-stage segmentation/detection framework. Empirical results on PANORAMA and an external cohort show that combining global features and parametric maps yields the best PDAC detection performance, with statistically significant gains over a baseline nnUNet and strong cross-dataset robustness. The approach is made scalable by CUDA-accelerated voxel-wise radiomics, enabling practical large-scale studies and potential clinical translation.

Abstract

Radiomics and deep learning both offer powerful tools for quantitative medical imaging, but most existing fusion approaches only leverage global radiomic features and overlook the complementary value of spatially resolved radiomic parametric maps. We propose a unified framework that first selects discriminative radiomic features and then injects them into a radiomics-enhanced nnUNet at both the global and voxel levels for pancreatic ductal adenocarcinoma (PDAC) detection. On the PANORAMA dataset, our method achieved AUC = 0.96 and AP = 0.84 in cross-validation. On an external in-house cohort, it achieved AUC = 0.95 and AP = 0.78, outperforming the baseline nnUNet; it also ranked second in the PANORAMA Grand Challenge. This demonstrates that handcrafted radiomics, when injected at both global and voxel levels, provide complementary signals to deep learning models for PDAC detection. Our code can be found at https://github.com/briandzt/dl-pdac-radiomics-global-n-paramaps

From Global Radiomics to Parametric Maps: A Unified Workflow Fusing Radiomics and Deep Learning for PDAC Detection

TL;DR

This work addresses PDAC detection by uniting handcrafted radiomics with deep learning through a radiomics-aware, two-stage nnUNet. The pipeline first selects a compact set of discriminative global radiomic features from the whole pancreas and then fuses these features, along with voxel-wise radiomic maps, into a two-stage segmentation/detection framework. Empirical results on PANORAMA and an external cohort show that combining global features and parametric maps yields the best PDAC detection performance, with statistically significant gains over a baseline nnUNet and strong cross-dataset robustness. The approach is made scalable by CUDA-accelerated voxel-wise radiomics, enabling practical large-scale studies and potential clinical translation.

Abstract

Radiomics and deep learning both offer powerful tools for quantitative medical imaging, but most existing fusion approaches only leverage global radiomic features and overlook the complementary value of spatially resolved radiomic parametric maps. We propose a unified framework that first selects discriminative radiomic features and then injects them into a radiomics-enhanced nnUNet at both the global and voxel levels for pancreatic ductal adenocarcinoma (PDAC) detection. On the PANORAMA dataset, our method achieved AUC = 0.96 and AP = 0.84 in cross-validation. On an external in-house cohort, it achieved AUC = 0.95 and AP = 0.78, outperforming the baseline nnUNet; it also ranked second in the PANORAMA Grand Challenge. This demonstrates that handcrafted radiomics, when injected at both global and voxel levels, provide complementary signals to deep learning models for PDAC detection. Our code can be found at https://github.com/briandzt/dl-pdac-radiomics-global-n-paramaps
Paper Structure (16 sections, 2 figures, 2 tables)

This paper contains 16 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Radiomics-Enhanced Deep Learning
  4. Application of Radiomics Parametric Maps
  5. Deep Learning Based Pancreatic Cancer Detection
  6. Method
  7. Data Preparation
  8. Global Radiomics Analysis
  9. Two-stage nnUNet for PDAC detection
  10. Experimental Results
  11. Global Radiomics-Based PDAC Classification
  12. PDAC Detection
  13. Runtime of Voxel-Wise Radiomics
  14. Conclusion
  15. ACKNOWLEDGMENTS
  16. ...and 1 more sections

Figures (2)

  • Figure 1: (Step 1) Whole-pancreas radiomics are extracted for PDAC classification, and the most informative features are selected. (Step 2) A coarse-to-fine two-stage network: Stage I predicts a rough mask to define the ROI and compute global/local radiomics; Stage II concatenates radiomics parametric maps with CT and fuses global radiomics via latent multi-head cross-attention to produce the final PDAC segmentation/detection.
  • Figure 2: ROC and PR curves for in-house external dataset PDAC detection.Our model with combined global and local radiomics features achieves better performance (0.662 -> 0.777, p<0.01) than the baseline nnUNet model.