Related Experiment Video
Updated: Jun 23, 2025
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
Presurgical Upgrade Prediction of DCIS to Invasive Ductal Carcinoma Using Time-dependent Deep Learning Models with
John D Mayfield1, Dana Ataya1, Mahmoud Abdalah1
1From the Departments of Radiology (J.D.M.), Oncologic Sciences (D.A., M.M.B., N.R., B.N.), and Medical Engineering (J.D.M.), University of South Florida College of Medicine, 12901 Bruce B. Downs Blvd, Tampa, FL 33612; and Department of Diagnostic Imaging and Interventional Radiology (D.A., B.N.), Department of Pathology (M.M.B.), Department of Cancer Physiology (N.R.), Quantitative Imaging CORE (M.A., O.S., I.E.N.), and Department of Machine Learning (M.M.B., I.E.N.), H. Lee Moffitt Cancer Center and Research Institute, Tampa, Fla.
Time-dependent deep learning models accurately predict ductal carcinoma in situ (DCIS) upgrade to invasive breast cancer using dynamic contrast-enhanced MRI. These models outperform single time point approaches without requiring lesion segmentation.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Ductal carcinoma in situ (DCIS) is a non-invasive breast condition with a risk of progressing to invasive malignancy.
- Accurate prediction of DCIS upgrade is crucial for appropriate treatment decisions.
- Dynamic Contrast-Enhanced (DCE) MRI provides valuable functional information for breast lesion characterization.
Purpose of the Study:
- To evaluate the efficacy of time-dependent deep learning models in predicting the preoperative upgrade of DCIS to invasive breast cancer.
- To compare the performance of sequential deep learning models against single time point models.
- To assess if lesion segmentation is a prerequisite for accurate prediction using these models.
Main Methods:
- Retrospective analysis of 154 preoperative DCE MRI scans from patients with biopsy-proven DCIS.
- Implementation of binary classification using Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) architectures.
- Benchmarking CNN-LSTM models against traditional CNNs, utilizing VGG16 and ResNet50, without manual lesion segmentation.
Main Results:
- VGG16-based CNN-LSTM models achieved higher multiphase test AUC (0.73) compared to ResNet50-based CNN-LSTM models (0.62).
- Time-dependent CNN-LSTM models demonstrated superior performance (AUC, 0.73) over single time point CNN models (AUC, 0.67).
- The developed deep learning models predicted DCIS upgrade without requiring manual lesion segmentation.
Conclusions:
- Sequential deep learning algorithms applied to preoperative DCE MRI can effectively predict the upgrade of DCIS to invasive malignancy.
- Time-dependent models offer improved predictive performance over single time point models for this task.
- These findings suggest a potential for automated, segmentation-free prediction of DCIS upgrade, aiding clinical decision-making.
