Breast lesion detection from DCE-MRI using YOLOv7


KULAVUZ B., Cavusoglu M., BAYRAM B., BAKIRMAN T., Sahin S., Araz N., ...More

International Conference of Computational Methods in Sciences and Engineering 2022, ICCMSE 2022, Hybrid, Heraklion, Greece, 26 - 29 October 2022, vol.3030 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 3030
  • Doi Number: 10.1063/5.0193021
  • City: Hybrid, Heraklion
  • Country: Greece
  • Yıldız Technical University Affiliated: Yes

Abstract

Breast cancer is one of the most common types of cancer among women. Early diagnosis of breast cancer has vital importance to prevent unexpected losses. A worldwide effort has been made to tackle early detection challenge. Dynamic contrast-enhanced magnetic resonance imaging is a superior imaging system that improves breast cancer diagnosis quality of physicians. Computer Aided Diagnosis systems are used as a complementary tool to improve breast cancer diagnosis. In last decades, various computer aided diagnosis systems have been proposed. However, the state-of-the-art deep learning-based approaches have started to overcome conventional medical image processing methods. In this study, we aimed to detect malignant breast lesions from open access dynamic contrast-enhanced magnetic resonance imagery dataset using most recent YOLOv7 deep learning architecture. 2400 images have been used for training (80%) and testing (20%) of the network. The metrics calculated with the test dataset are 98.54%, 96.42% and 84.40% for mAP@0.50 IoU, mAP@0.75 IoU and mAP, respectively. The results show that YOLOv7 architecture is capable to detect malignant breast lesions from dynamic contrast-enhanced magnetic resonance images efficiently.