Radiomic analysis in contrast-enhanced mammography using a multivendor data set: accuracy of models according to...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Radiomic analysis of contrast-enhanced mammography (CEM) is an emerging field.
• Developing accurate classification models is crucial for distinguishing benign from malignant lesions.

Purpose of the Study:

• To build classification models distinguishing benign and malignant lesions using a multivendor CEM dataset.
• To compare the performance of different segmentation techniques (freehand ROI vs. ellipsoid_ROI).

Main Methods:

• CEM images from Hologic and GE equipment were analyzed.
• Textural features were extracted using MaZda software.
• Lesions were segmented using freehand region of interest (ROI) and ellipsoid_ROI.

Main Results:

• High diagnostic accuracy (>0.9) was achieved for all models.
• Ellipsoid_ROI segmentation resulted in a more accurate model (accuracy: 0.947, AUC: 0.974) compared to FH_ROI (accuracy: 0.914, AUC: 0.86).
• Models using single mammographic views (CC or MLO) showed high accuracy, with the CC-view model having the highest specificity.

Conclusions:

• Accurate radiomics models can be developed from real-world, multivendor CEM data.
• Ellipsoid_ROI segmentation provides superior accuracy for radiomic analysis.
• Segmenting both mammographic views may not be necessary, potentially reducing workload.