Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

STA-Risk: A Deep Dive of Spatio-Temporal Asymmetries for Breast Cancer Risk Prediction

Zhengbo Zhou, Dooman Arefan, Margarita Zuley, Jules Sumkin, Shandong Wu

TL;DR

STA-Risk tackles the challenge of predicting breast cancer risk from longitudinal mammograms by modeling spatial and temporal tissue asymmetries. It introduces a Transformer-based framework with a learnable side embedding and continuous-time temporal encoding, regulated by an asymmetry loss, and uses an additive-hazard-style risk predictor. Across two independent datasets, STA-Risk outperforms state-of-the-art baselines on 1- to 5-year horizons, demonstrated by higher C-index and AUC metrics. The approach highlights the clinical value of leveraging bilateral and longitudinal imaging cues for personalized risk stratification and screening, with plans to release source code. $\text{Risk}(k) = \beta_0 + \sum_{j=1}^k \beta_j$ captures cumulative horizon risk, and the asymmetry loss reinforces predictive sensitivity to clinically meaningful left-right and temporal changes.$

Abstract

Predicting the risk of developing breast cancer is an important clinical tool to guide early intervention and tailoring personalized screening strategies. Early risk models have limited performance and recently machine learning-based analysis of mammogram images showed encouraging risk prediction effects. These models however are limited to the use of a single exam or tend to overlook nuanced breast tissue evolvement in spatial and temporal details of longitudinal imaging exams that are indicative of breast cancer risk. In this paper, we propose STA-Risk (Spatial and Temporal Asymmetry-based Risk Prediction), a novel Transformer-based model that captures fine-grained mammographic imaging evolution simultaneously from bilateral and longitudinal asymmetries for breast cancer risk prediction. STA-Risk is innovative by the side encoding and temporal encoding to learn spatial-temporal asymmetries, regulated by a customized asymmetry loss. We performed extensive experiments with two independent mammogram datasets and achieved superior performance than four representative SOTA models for 1- to 5-year future risk prediction. Source codes will be released upon publishing of the paper.

STA-Risk: A Deep Dive of Spatio-Temporal Asymmetries for Breast Cancer Risk Prediction

TL;DR

STA-Risk tackles the challenge of predicting breast cancer risk from longitudinal mammograms by modeling spatial and temporal tissue asymmetries. It introduces a Transformer-based framework with a learnable side embedding and continuous-time temporal encoding, regulated by an asymmetry loss, and uses an additive-hazard-style risk predictor. Across two independent datasets, STA-Risk outperforms state-of-the-art baselines on 1- to 5-year horizons, demonstrated by higher C-index and AUC metrics. The approach highlights the clinical value of leveraging bilateral and longitudinal imaging cues for personalized risk stratification and screening, with plans to release source code. captures cumulative horizon risk, and the asymmetry loss reinforces predictive sensitivity to clinically meaningful left-right and temporal changes.$

Abstract

Predicting the risk of developing breast cancer is an important clinical tool to guide early intervention and tailoring personalized screening strategies. Early risk models have limited performance and recently machine learning-based analysis of mammogram images showed encouraging risk prediction effects. These models however are limited to the use of a single exam or tend to overlook nuanced breast tissue evolvement in spatial and temporal details of longitudinal imaging exams that are indicative of breast cancer risk. In this paper, we propose STA-Risk (Spatial and Temporal Asymmetry-based Risk Prediction), a novel Transformer-based model that captures fine-grained mammographic imaging evolution simultaneously from bilateral and longitudinal asymmetries for breast cancer risk prediction. STA-Risk is innovative by the side encoding and temporal encoding to learn spatial-temporal asymmetries, regulated by a customized asymmetry loss. We performed extensive experiments with two independent mammogram datasets and achieved superior performance than four representative SOTA models for 1- to 5-year future risk prediction. Source codes will be released upon publishing of the paper.

Paper Structure

This paper contains 12 sections, 6 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Overall Network Architecture
  4. Side Encoding
  5. Temporal Encoding
  6. Asymmetry Loss
  7. Experiments
  8. Study Cohorts and Datasets
  9. Implementation Details
  10. Results
  11. Conclusion
  12. Acknowledgments

Figures (2)

  • Figure 1: The proposed STA-Risk architecture that aims to capture spatial and temporal asymmetry from longitudinal screening mammogram exams for predicting breast cancer risk. The key components include side-aware spatial encoding, temporal attention, and a customized asymmetry loss.
  • Figure 2: (Left) Visualization on the Differences when Using STA-Risk vs. Without Side Encoding and Asymmetric Loss. (Right) Illustrative Visualization on the Effects of the STA-Risk Model.