Cancer-Net PCa-MultiSeg: Multimodal Enhancement of Prostate Cancer Lesion Segmentation Using Synthetic Correlated Diffusion Imaging
Jarett Dewbury, Chi-en Amy Tai, Alexander Wong · 2025
Read Cancer-Net PCa-MultiSeg: Multimodal Enhancement of Prostate Cancer Lesion Segmentation Using Synthetic Correlated Diffusion Imaging by Jarett Dewbury, Chi-en Amy Tai, Alexander Wong (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-07816.txt.
Introduction
Jarett Dewbury, Chi-en Amy Tai, and Alexander Wong sharpen prostate cancer maps with a synthetic imaging channel. Deep learning segmentation of lesions stalls at low overlap scores in large patient groups. Synthetic correlated diffusion imaging adds information beyond standard diffusion scans, and the team tests whether it lifts performance.
Six leading segmentation architectures run on two hundred patients with aligned synthetic, diffusion-weighted, and coefficient maps. Adding the synthetic channel improved or preserved results in nearly all configurations, with some architectures gaining strongly. Better lesion outlines support biopsy targeting and treatment planning. Multimodal gains from a computed channel cost no extra scan time.
Catalog record
| CREATOR | Jarett Dewbury, Chi-en Amy Tai, Alexander Wong |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 16 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-07816
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Cancer-Net PCa-MultiSeg: Multimodal Enhancement of Prostate Cancer Lesion Segmentation Using Synthetic Correlated Diffusion Imaging Jarett Dewbury, Chi-en Amy Tai, Alexander Wong 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. † † workshop: Machine Learning for Health (ML4H) 2025 keywords prostate cancer, medical image segmentation, synthetic correlated diffusion imaging, multiparametric MRI, deep learning Data and Code Availability This study leveraged the Cancer-Net PCa-Data data set ( 5 ) , a publicly available data set designed for multimodal PCa analysis. The data set was derived from the SPIE-AAPM-NCI PROSTATEx Challenges, PROSTATEx_masks, and The Cancer Imaging Archive (TCIA) datasets ( 8 ; 9 ; 3 ; 2 ) . Code is available at https://github.com/Jdewbury/Cancer-Net-PCa-MultiSeg Institutional Review Board (IRB) This research does not require IRB approval, as it exclusively uses publicly available data that have previously been approved for research use. 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. 1 Introduction Prostate cancer (PCa) is the leading cancer diagnosis among men in the United States, accounting for more than 310,000 new cases and 35,000 total deaths in 2025 ( 14 ) . Accurate localization and detection of PCa lesions remain fundamental in guiding treatment protocols and improving patient outcomes. Multiparametric magnetic resonance imaging (mpMRI) has established itself as the primary diagnostic modality for the evaluation of PCa, allowing non-invasive visualization of prostatic tissue and serving as the basis for standardized assessment protocols such as the Prostate Imaging Reporting and Data System (PI-RADS) ( 15 ) . Despite technological advances, current automated approaches for PCa lesion detection exhibit significant performance constraints. Deep learning methods applied to mpMRI data for lesion segmentation demonstrate limited efficacy, with recent investigations reporting Dice score values of 0.32 ( 4 ) and 0.276 ( 12 ) on large patient cohorts. These results highlight the persistent challenges of accurately differentiating malignant and benign prostatic tissue, emphasizing the need for improved imaging techniques that offer enhanced diagnostic precision. However, such approaches must also maintain clinical feasibility by balancing gains in diagnostic accuracy with reasonable computational demands and minimal modifications to established workflows. Synthetic correlated diffusion imaging (CDI s ) presents a promising approach to address current limitations in PCa lesion detection. CDI s leverages existing diffusion-weighted acquisitions to enhance tissue contrast while maintaining compatibility with established mpMRI protocols and requiring no additional scan time. This technique demonstrates superior delineation between cancerous and healthy prostatic tissue, as well as fewer false positives compared to standard mpMRI techniques ( 16 ) . In this study, we investigate the integration of CDI s with standard diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC), which constitute the diffusion component of contemporary mpMRI protocols. DWI serves as the dominant sequence for peripheral zone assessment in the PI-RADS evaluation, while ADC provides complementary diffusion characterization ( 15 ) . The systematic evaluation of CDI s enhancement in various deep learning architectures for PCa lesion segmentation remains unexplored. This work presents the first comprehensive benchmark evaluating CDI s integration with standard DWI and ADC sequences in six state-of-the-art medical segmentation models, demonstrating improved lesion detection performance and establishing a framework for integrating advanced imaging techniques into existing clinical workflows to enhance PCa diagnostic capabilities. 2 Methodology 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. 2.1 Dataset This study uses the Cancer-Net PCa-Data data set comprising volumetric imaging data from a cohort of 200 patients, including three spatially co-registered diffusion-based sequences: CDI s , DWI and ADC (see Figure 1 for sample patient data). These modalities share identical spatial coordinates and voxel dimensions, allowing for direct pixel-wise comparison and eliminating registration-related artifacts that could confound multimodal analysis. Ground truth annotations include lesion masks validated by expert radiologists, encompassing both clinically significant and insignificant PCa lesions defined by Gleason scoring criteria. CDI s Computation. CDI s is derived from standard DWI acquisitions without requiring additional scan time ( 16 ) . The computation involves three steps: (1) multiple DWI signals are acquired at different b-values; (2) additional signals at higher b-values are synthesized from native acquisitions using a signal synthesizer, avoiding lengthy high b-value acquisitions that would increase scan time and introduce artifacts; (3) native and synthetic signals are combined through a correlation-based mixing function that analyzes signal intensity co-variation across different b-values. This process preserves information on the distribution of water mobility within cancerous regions that exhibit high signal across all b-values, while healthy tissue shows high signal only at low b-values. Unlike ADC, which reduces diffusion to a single number, CDI s retains distributional information to enable better differentiation of malignant tissue. Since CDI s utilizes existing DWI sequences, it integrates seamlessly into clinical workflows without protocol modifications. 2.2 Experimental Setup Our experimental pipeline (Figure 2 ) facilitates a systematic evaluation of the enhancement of CDI s in multiple architectures and combinations of modalities. Patient-level stratification was applied to prevent data leakage, with 90% allocated to training/validation and 10% to the held-out test set. The training/validation subset was further partitioned using 10-fold cross-validation, enabling robust performance estimation across diverse patient populations. All modality and ground-truth lesion mask volume pairs were standardized to 128x128x19 voxels using bilinear interpolation for images and nearest-neighbor interpolation for masks. Channel-wise min-max normalization was applied to the modality volumes to ensure uniform intensity scaling in the [0, 1] range. Seven modality configurations were systematically evaluated to assess CDI s enhancement across three categories: individual modality (CDI s , ADC, DWI), standard clinical protocol (DWI + ADC), and CDI s -enhanced combinations (CDI s + ADC, CDI s + DWI, CDI s + DWI + ADC). This design enables direct assessment of CDI s performance contributions when integrated with standard diffusion sequences while establishing individual modality baselines for PCa lesion segmentation. Modality U-Net Attention U-Net UNet++ SegResNet UNETR Swin UNETR CDI s 24.64\pm 3.1 26.72\pm 1.4 27.27\pm 1.5 27.58\pm 1.4 26.34\pm 1.3 26.86\pm 3.1 ADC 17.47\pm 3.2 25.50\pm 4.0 31.58\pm 1.7 30.74\pm 3.1