Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cancer-Net PCa-MultiSeg: Multimodal Enhancement of Prostate Cancer Lesion Segmentation Using Synthetic Correlated Diffusion Imaging

Jarett Dewbury, Chi-en Amy Tai, Alexander Wong

TL;DR

The paper tackles the persistent difficulty of automated prostate cancer lesion segmentation, where Dice scores have ranged as low as $0.32$ in large cohorts. It introduces synthetic correlated diffusion imaging ($CDI^s$) as a zero-scan-time enhancement to diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC), and benchmarks its impact across six architectures on a 200-patient multimodal dataset. The results show that CDI$^s$ integration improves or preserves Dice scores in $94\%$ of configurations, with up to $+72.5\%$ relative improvements (notably for Attention U‑Net with CDI$^s$+DWI) and $+5.13$ as a maximum absolute gain for U‑Net; CDI$^s$+DWI is the safest configuration with zero degradations in three architectures. The work provides deployment-ready guidance, demonstrating that CDI$^s$ can be deployed as a practical drop-in enhancement without altering clinical workflows, and suggests directions for future optimization on transformer-based architectures and broader dataset validation.

Abstract

Current deep learning approaches for prostate cancer lesion segmentation achieve limited performance, with Dice scores of 0.32 or lower in large patient cohorts. To address this limitation, we investigate synthetic correlated diffusion imaging (CDI$^s$) as an enhancement to standard diffusion-based protocols. We conduct a comprehensive evaluation across six state-of-the-art segmentation architectures using 200 patients with co-registered CDI$^s$, diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) sequences. We demonstrate that CDI$^s$ integration reliably enhances or preserves segmentation performance in 94% of evaluated configurations, with individual architectures achieving up to 72.5% statistically significant relative improvement over baseline modalities. CDI$^s$ + DWI emerges as the safest enhancement pathway, achieving significant improvements in half of evaluated architectures with zero instances of degradation. Since CDI$^s$ derives from existing DWI acquisitions without requiring additional scan time or architectural modifications, it enables immediate deployment in clinical workflows. Our results establish validated integration pathways for CDI$^s$ as a practical drop-in enhancement for PCa lesion segmentation tasks across diverse deep learning architectures.

Cancer-Net PCa-MultiSeg: Multimodal Enhancement of Prostate Cancer Lesion Segmentation Using Synthetic Correlated Diffusion Imaging

TL;DR

The paper tackles the persistent difficulty of automated prostate cancer lesion segmentation, where Dice scores have ranged as low as in large cohorts. It introduces synthetic correlated diffusion imaging () as a zero-scan-time enhancement to diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC), and benchmarks its impact across six architectures on a 200-patient multimodal dataset. The results show that CDI integration improves or preserves Dice scores in of configurations, with up to relative improvements (notably for Attention U‑Net with CDI+DWI) and as a maximum absolute gain for U‑Net; CDI+DWI is the safest configuration with zero degradations in three architectures. The work provides deployment-ready guidance, demonstrating that CDI can be deployed as a practical drop-in enhancement without altering clinical workflows, and suggests directions for future optimization on transformer-based architectures and broader dataset validation.

Abstract

Current deep learning approaches for prostate cancer lesion segmentation achieve limited performance, with Dice scores of 0.32 or lower in large patient cohorts. To address this limitation, we investigate synthetic correlated diffusion imaging (CDI) as an enhancement to standard diffusion-based protocols. We conduct a comprehensive evaluation across six state-of-the-art segmentation architectures using 200 patients with co-registered CDI, diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) sequences. We demonstrate that CDI integration reliably enhances or preserves segmentation performance in 94% of evaluated configurations, with individual architectures achieving up to 72.5% statistically significant relative improvement over baseline modalities. CDI + DWI emerges as the safest enhancement pathway, achieving significant improvements in half of evaluated architectures with zero instances of degradation. Since CDI derives from existing DWI acquisitions without requiring additional scan time or architectural modifications, it enables immediate deployment in clinical workflows. Our results establish validated integration pathways for CDI as a practical drop-in enhancement for PCa lesion segmentation tasks across diverse deep learning architectures.

Paper Structure

This paper contains 8 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dataset
  4. Experimental Setup
  5. Model Architectures
  6. Training Protocol
  7. Results
  8. Conclusion

Figures (2)

  • Figure 1: Visualization of sample T2-weighted (T2w) images with overlays highlighting lesion boundaries (red dashed lines) and comparative intensity distributions for CDI$^s$, DWI, and ADC sequences. We consider both clinically significant (csPCa) and clinically insignificant (insPCa) lesions.
  • Figure 2: Multimodal prostate cancer lesion segmentation pipeline. The workflow includes modular input volume selection from three modalities (CDI$^s$, DWI, ADC), standardized preprocessing, evaluation across six segmentation architectures, and binary lesion mask prediction output.