Scholarlink

Accurate identification of the prostate gland on magnetic resonance imaging (MRI) is a critical step in the diagnostic and treatment planning workflow for prostate cancer. Manual delineation, however, remains time-consuming and prone to interobserver variability. This study investigates the applicability of YOLOv8, a state-of-the-art real-time object detection model, for automatic prostate gland detection in axial T2-weighted MRI slices. A dataset of 71 patients consisting of 1019 annotated prostate slices were used, encompassing heterogeneous imaging conditions and glandular anatomies. YOLOv8 was trained using advanced preprocessing techniques, anchor-free detection, and optimized loss functions to improve localization precision and inference speed. The model achieved a mean Average Precision (mAP@0.5) of 0.98, with peak precision and recall values of 1.00 and 0.99, respectively. The F1-score reached 0.97, indicating a strong balance between detection accuracy and sensitivity. Although the confusion matrix revealed a mild obstacle in background separation, the overall results demonstrate that YOLOv8 offers robust, efficient, and clinically promising performance for automated prostate gland detection, facilitating faster and more consistent MRI interpretation.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated Prostate Gland Detection in MRI Using YOLOv8: A Deep Learning Approach for Early Diagnosis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Akhtar Rasool,
  • Muhammad Ali,
  • Leonardo Salvaggio,
  • Viviana Benfante,
  • Albert Comelli

摘要

Accurate identification of the prostate gland on magnetic resonance imaging (MRI) is a critical step in the diagnostic and treatment planning workflow for prostate cancer. Manual delineation, however, remains time-consuming and prone to interobserver variability. This study investigates the applicability of YOLOv8, a state-of-the-art real-time object detection model, for automatic prostate gland detection in axial T2-weighted MRI slices. A dataset of 71 patients consisting of 1019 annotated prostate slices were used, encompassing heterogeneous imaging conditions and glandular anatomies. YOLOv8 was trained using advanced preprocessing techniques, anchor-free detection, and optimized loss functions to improve localization precision and inference speed. The model achieved a mean Average Precision (mAP@0.5) of 0.98, with peak precision and recall values of 1.00 and 0.99, respectively. The F1-score reached 0.97, indicating a strong balance between detection accuracy and sensitivity. Although the confusion matrix revealed a mild obstacle in background separation, the overall results demonstrate that YOLOv8 offers robust, efficient, and clinically promising performance for automated prostate gland detection, facilitating faster and more consistent MRI interpretation.