Development of a YOLOv8-Based Model for Classification of Contrast-Enhanced Mammograms with Receptor Integration

Avcı Yön H., Karakaya Karabulut J., Durhan G., Gülsün Akpınar M., Demirkazık F.

Joint Conference of the Italian and Eastern Mediterranean Regions of the International Biometric Society IBS-IR-EMR2025, Salerno, Italy, 16 - 18 September 2025, pp.231, (Summary Text)

  • Publication Type: Conference Paper / Summary Text
  • City: Salerno
  • Country: Italy
  • Page Numbers: pp.231
  • Hacettepe University Affiliated: Yes

Abstract

Background: Contrast-enhanced mammography (CEM) is a promising modality for detecting and characterizing breast lesions. While deep learning models have shown high performance in image-based classification tasks, few studies have incorporated biological markers to enhance diagnostic accuracy. The integration of hormone receptor status—specifically estrogen receptor (ER), progesterone receptor (PR), and HER2—remains largely underexplored in the context of CEM image analysis.


Objective: This study aims to develop a deep learning-based model utilizing the YOLOv8 architecture for the automated segmentation and classification of CEM images into three categories: normal, benign, and malignant. A key novelty of this work lies in the integration of ER, PR, and HER2 receptor status—derived from pathology reports—into the model to improve classification performance.


Methods: CEM images were obtained from the Radiology Department of Hacettepe University, under ethical approval from the Non-Interventional Clinical Research Ethics Committee. A YOLOv8 segmentation-based model is being trained to automatically detect and localize lesions. Hormone receptor status will be fused with the image-based features through a multi-modal learning pipeline. Preliminary stages, including preprocessing, annotation, and model training, are currently ongoing. Model performance will be evaluated using classification accuracy, AUC, and F1-score, with and without receptor integration.


Conclusion: To the best of our knowledge, this is one of the first studies to incorporate YOLOv8-based segmentation with hormone receptor information for multiclass classification of contrast-enhanced mammograms. This integrated approach is expected to enhance diagnostic performance by combining radiological morphology with tumor biology, contributing to more personalized and biologically informed breast cancer diagnostics.


Keywords: Contrast-Enhanced Mammography, Convolutional Neural Network, Hormone Receptor Integration