Related Experiment Video
Updated: Aug 15, 2025
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
Differences in Molecular Subtype Reference Standards Impact AI-based Breast Cancer Classification with Dynamic
Yu Ji1, Heather M Whitney1, Hui Li1
1From the Department of Radiology, The Second Hospital of Tianjin Medical University, No. 23 Pingjiang Rd, Hexi District, Tianjin, China 300211 (Y.J., X.Z.); National Clinical Research Center for Cancer, Tianjin Medical University Cancer Institute and Hospital, Tianjin, China (Y.J., P.L.); Department of Radiology, The University of Chicago, Chicago, Ill (H.M.W., H.L., M.L.G.); and Department of Physics, Wheaton College, Wheaton, Ill (H.M.W.).
Discrepancies in breast cancer subtype classification impact AI performance on MRI scans. Using radiomic features from dynamic contrast-enhanced MRI shows potential for improved classification accuracy when reference standards agree.
Area of Science:
- Radiology
- Oncology
- Artificial Intelligence
Background:
- Breast cancer molecular subtypes (luminal A/B) are defined by immunohistochemical (IHC) staining or St Gallen criteria.
- Disagreement exists between IHC and St Gallen criteria for classifying these subtypes.
Purpose of the Study:
- To investigate the impact of reference standard disagreement on AI classification of luminal A and B breast cancer subtypes.
- To analyze radiomic features from dynamic contrast-enhanced (DCE) MRI for subtype classification.
Main Methods:
- Retrospective analysis of 877 breast tumors using 28 radiomic features from DCE-MRI.
- Comparison of radiomic features between tumors with agreement and disagreement in IHC and St Gallen classifications.
- AI classification using linear discriminant analysis and stepwise feature selection with fivefold cross-validation.
Main Results:
- Six radiomic features differed significantly (P ≤ .001) between agreement and disagreement groups.
- AI classification achieved a higher area under the receiver operating characteristic curve (AUC) of 0.74 when using tumors with agreement between reference standards, compared to using individual standards (IHC AUC: 0.66, St Gallen AUC: 0.62).
Conclusions:
- Discrepancies in molecular subtype reference standards negatively affect AI classification performance for luminal breast cancer subtypes using DCE-MRI.
- Standardized classification criteria are crucial for reliable AI-driven radiomic analysis in breast cancer.
More Related Videos
06:24Dynamic Contrast Enhanced Magnetic Resonance Imaging of an Orthotopic Pancreatic Cancer Mouse Model
Published on: April 18, 2015
09:53Quantifying the Brain Metastatic Tumor Micro-Environment using an Organ-On-A Chip 3D Model, Machine Learning, and Confocal Tomography
Published on: August 16, 2020