Related Experiment Video
Updated: Oct 19, 2025
08:05
Detection and Isolation of Cancer in Prostate Biopsies Using Stimulated Raman Histology and Artificial Intelligence
Published on: June 10, 2025
776
Automatic Detection and Tracking of Marker Seeds Implanted in Prostate Cancer Patients using a Deep Learning
Keya Amarsee1, Prabhakar Ramachandran1,2,3, Andrew Fielding2
1Department of Radiation Oncology, Princess Alexandra Hospital, Woolloongabba, Australia.
Journal of Medical Physics
|September 27, 2021
Summary
This study demonstrates deep learning using You Only Look Once (YOLO) v2 for real-time detection of prostate cancer fiducial marker seeds in X-ray images. The method achieved 98% detection accuracy, aiding precise radiation therapy delivery.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Radiation Oncology
Background:
- Fiducial marker seeds are crucial for tracking prostate volume during cancer treatment.
- Accurate seed tracking minimizes radiation dose spillage, reducing normal tissue complications.
Purpose of the Study:
- To develop and validate a real-time system for detecting fiducial marker seeds using deep learning.
- To utilize the You Only Look Once (YOLO) v2 convolutional neural network for seed detection on kilovoltage X-ray volume imaging (kV XVI) panel images.
- To create a MATLAB-based program for visualizing and detecting seeds in projection images.
Main Methods:
- A You Only Look Once (YOLO) v2 model was trained using labeled ground truth datasets of fiducial marker seeds.
- Kilovoltage X-ray volume imaging (kV XVI) panel images from a wax phantom and three patients were used for training and testing.
- A software program was developed to display projection images in real-time, predict seed locations using YOLO v2, and determine seed centers.
Main Results:
- Fiducial marker seeds were detected with 98% accuracy across all gantry angles.
- The variation in detected seed center position was within ±1 mm.
- The percentage difference between ground truth and detected seeds was within 3%.
Conclusions:
- Deep learning, specifically YOLO v2, is effective for real-time fiducial marker seed detection in kV images.
- This technology has the potential to improve the accuracy and efficiency of image-guided radiation therapy.
- Future work aims to extend this approach for tracking moving structures in other anatomical sites.
