Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Second-Order Attention Mechanism For Prostate Cancer Segmentation and Detection in Bi-Parametric MRI

Mateo Ortiz, Juan Olmos, Fabio Martínez

TL;DR

This work tackles csPCa detection and segmentation in bp-MRI under limited annotated data by introducing a second-order geometric attention (SOGA) that operates on SPD descriptors within a Riemannian manifold to refine skip connections in U-Net and nnU-Net architectures. By compressing feature banks into SPD matrices, applying BiMap/ReEig/LogEig blocks, and fusing encoder/decoder geometry through attention coefficients, SOGA yields more discriminative representations for lesion detection. Empirical results on PI-CAI and PROSTATE158 show consistent improvements in AUC-ROC, AP, Sen@1FP, and DSC over baselines and first-order attention, with particularly strong generalization to unseen cohorts. While computational cost rises due to Riemannian operations, the approach demonstrates robust, lesion-size–dependent gains and offers a scalable path toward reliable clinical deployment in csPCa screening with bp-MRI.

Abstract

The detection of clinically significant prostate cancer lesions (csPCa) from biparametric magnetic resonance imaging (bp-MRI) has emerged as a noninvasive imaging technique for improving accurate diagnosis. Nevertheless, the analysis of such images remains highly dependent on the subjective expert interpretation. Deep learning approaches have been proposed for csPCa lesions detection and segmentation, but they remain limited due to their reliance on extensively annotated datasets. Moreover, the high lesion variability across prostate zones poses additional challenges, even for expert radiologists. This work introduces a second-order geometric attention (SOGA) mechanism that guides a dedicated segmentation network, through skip connections, to detect csPCa lesions. The proposed attention is modeled on the Riemannian manifold, learning from symmetric positive definitive (SPD) representations. The proposed mechanism was integrated into standard U-Net and nnU-Net backbones, and was validated on the publicly available PI-CAI dataset, achieving an Average Precision (AP) of 0.37 and an Area Under the ROC Curve (AUC-ROC) of 0.83, outperforming baseline networks and attention-based methods. Furthermore, the approach was evaluated on the Prostate158 dataset as an independent test cohort, achieving an AP of 0.37 and an AUC-ROC of 0.75, confirming robust generalization and suggesting discriminative learned representations.

A Second-Order Attention Mechanism For Prostate Cancer Segmentation and Detection in Bi-Parametric MRI

TL;DR

This work tackles csPCa detection and segmentation in bp-MRI under limited annotated data by introducing a second-order geometric attention (SOGA) that operates on SPD descriptors within a Riemannian manifold to refine skip connections in U-Net and nnU-Net architectures. By compressing feature banks into SPD matrices, applying BiMap/ReEig/LogEig blocks, and fusing encoder/decoder geometry through attention coefficients, SOGA yields more discriminative representations for lesion detection. Empirical results on PI-CAI and PROSTATE158 show consistent improvements in AUC-ROC, AP, Sen@1FP, and DSC over baselines and first-order attention, with particularly strong generalization to unseen cohorts. While computational cost rises due to Riemannian operations, the approach demonstrates robust, lesion-size–dependent gains and offers a scalable path toward reliable clinical deployment in csPCa screening with bp-MRI.

Abstract

The detection of clinically significant prostate cancer lesions (csPCa) from biparametric magnetic resonance imaging (bp-MRI) has emerged as a noninvasive imaging technique for improving accurate diagnosis. Nevertheless, the analysis of such images remains highly dependent on the subjective expert interpretation. Deep learning approaches have been proposed for csPCa lesions detection and segmentation, but they remain limited due to their reliance on extensively annotated datasets. Moreover, the high lesion variability across prostate zones poses additional challenges, even for expert radiologists. This work introduces a second-order geometric attention (SOGA) mechanism that guides a dedicated segmentation network, through skip connections, to detect csPCa lesions. The proposed attention is modeled on the Riemannian manifold, learning from symmetric positive definitive (SPD) representations. The proposed mechanism was integrated into standard U-Net and nnU-Net backbones, and was validated on the publicly available PI-CAI dataset, achieving an Average Precision (AP) of 0.37 and an Area Under the ROC Curve (AUC-ROC) of 0.83, outperforming baseline networks and attention-based methods. Furthermore, the approach was evaluated on the Prostate158 dataset as an independent test cohort, achieving an AP of 0.37 and an AUC-ROC of 0.75, confirming robust generalization and suggesting discriminative learned representations.

Paper Structure

This paper contains 15 sections, 2 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Method
  4. U-Net-based representation
  5. Geometric Riemannian learning
  6. Second Order Geometric Attention module (SOGA)
  7. Dataset
  8. PI-CAI challenge dataset.
  9. PROSTATE158.
  10. Experimental Setup
  11. Evaluation.
  12. Results
  13. Detection in PI-CAI Public Cohort
  14. Generalization evaluation
  15. Discussion and conclusions remarks

Figures (3)

  • Figure 1: Pipeline of the proposed sencod-order geometric attention (SOGA) module integrated in a U-Net network.
  • Figure 2: Datasets flowchart. The training and validation of models was performed on the public PI-CAI cohort. For generalization evaluation we considered the hidden PI-CAI Tuning cohort and the Prostate158 dataset.
  • Figure 3: Visualization of predictions for three cases from the Prostate158 dataset, each presented in one row. The first three columns show bp-MRI images with ground-truth annotations. The predictions include DSC values.