Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cancer-Net PCa-Seg: Benchmarking Deep Learning Models for Prostate Cancer Segmentation Using Synthetic Correlated Diffusion Imaging

Jarett Dewbury, Chi-en Amy Tai, Alexander Wong

TL;DR

This work addresses prostate cancer gland segmentation using a novel MRI modality, synthetic correlated diffusion imaging (CDI$^s$). It benchmarks five state-of-the-art segmentation models (U-Net, SegResNet, Swin UNETR, Attention U-Net, LightM-UNet) on a $200$-patient CDI$^s$ dataset (Cancer-Net PCa-Data) with volumes resized to $128\times128\times19$ and evaluated primarily by the Dice-Sørensen coefficient ($DSC$). SegResNet achieves the top test performance ($DSC = 76.68 \pm 0.8$), while Attention U-Net offers a strong accuracy-efficiency trade-off; other models show varying trade-offs in inference time and parameter counts. The study demonstrates the feasibility of CDI$^s$-driven prostate gland segmentation and provides guidance on model selection for clinical deployment, while highlighting the need to validate generalizability across additional CDI$^s$ cohorts in future work.

Abstract

Prostate cancer (PCa) is the most prevalent cancer among men in the United States, accounting for nearly 300,000 cases, 29\% of all diagnoses and 35,000 total deaths in 2024. Traditional screening methods such as prostate-specific antigen (PSA) testing and magnetic resonance imaging (MRI) have been pivotal in diagnosis, but have faced limitations in specificity and generalizability. In this paper, we explore the potential of enhancing PCa gland segmentation using a novel MRI modality called synthetic correlated diffusion imaging (CDI$^s$). We employ several state-of-the-art deep learning models, including U-Net, SegResNet, Swin UNETR, Attention U-Net, and LightM-UNet, to segment prostate glands from a 200 CDI$^s$ patient cohort. We find that SegResNet achieved superior segmentation performance with a Dice-Sorensen coefficient (DSC) of $76.68 \pm 0.8$. Notably, the Attention U-Net, while slightly less accurate (DSC $74.82 \pm 2.0$), offered a favorable balance between accuracy and computational efficiency. Our findings demonstrate the potential of deep learning models in improving prostate gland segmentation using CDI$^s$ to enhance PCa management and clinical support.

Cancer-Net PCa-Seg: Benchmarking Deep Learning Models for Prostate Cancer Segmentation Using Synthetic Correlated Diffusion Imaging

TL;DR

This work addresses prostate cancer gland segmentation using a novel MRI modality, synthetic correlated diffusion imaging (CDI). It benchmarks five state-of-the-art segmentation models (U-Net, SegResNet, Swin UNETR, Attention U-Net, LightM-UNet) on a -patient CDI dataset (Cancer-Net PCa-Data) with volumes resized to and evaluated primarily by the Dice-Sørensen coefficient (). SegResNet achieves the top test performance (), while Attention U-Net offers a strong accuracy-efficiency trade-off; other models show varying trade-offs in inference time and parameter counts. The study demonstrates the feasibility of CDI-driven prostate gland segmentation and provides guidance on model selection for clinical deployment, while highlighting the need to validate generalizability across additional CDI cohorts in future work.

Abstract

Prostate cancer (PCa) is the most prevalent cancer among men in the United States, accounting for nearly 300,000 cases, 29\% of all diagnoses and 35,000 total deaths in 2024. Traditional screening methods such as prostate-specific antigen (PSA) testing and magnetic resonance imaging (MRI) have been pivotal in diagnosis, but have faced limitations in specificity and generalizability. In this paper, we explore the potential of enhancing PCa gland segmentation using a novel MRI modality called synthetic correlated diffusion imaging (CDI). We employ several state-of-the-art deep learning models, including U-Net, SegResNet, Swin UNETR, Attention U-Net, and LightM-UNet, to segment prostate glands from a 200 CDI patient cohort. We find that SegResNet achieved superior segmentation performance with a Dice-Sorensen coefficient (DSC) of . Notably, the Attention U-Net, while slightly less accurate (DSC ), offered a favorable balance between accuracy and computational efficiency. Our findings demonstrate the potential of deep learning models in improving prostate gland segmentation using CDI to enhance PCa management and clinical support.
Paper Structure (7 sections, 2 figures, 1 table)

This paper contains 7 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dataset
  4. Data Preparation
  5. Training
  6. Results
  7. Conclusion

Figures (2)

  • Figure 1: Visualization of sample data from the patient cohort in Cancer-Net PCa-Data dataset. T2-weighted (T2w) images are shown with CDI$^s$ prostate gland region boundary in color overlay. For the purposes of our paper, we consider both clinically significant (csPCa) and clinically insignificant (insPCa) prostate cancer segmentation.
  • Figure 2: Prostate cancer segmentation pipeline using Cancer-Net PCa-Data. The workflow includes (left to right): input CDI$^s$ volume slice, preprocessing, segmentation across five different deep learning models, and the final predicted prostate gland mask output.