Using a deep learning approach for implanted seed detection on fluoroscopy images in prostate brachytherapy

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Prostate brachytherapy involves implanting radioactive seeds for cancer treatment.
• Accurate localization of these seeds is crucial for effective treatment planning and delivery.
• Manual seed detection on fluoroscopy images can be time-consuming and prone to error.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated detection of implanted seeds in prostate brachytherapy fluoroscopy images.
• To improve the efficiency and accuracy of seed localization in clinical practice.

Main Methods:

• Utilized 48 fluoroscopy images from patients undergoing permanent seed implant (PSI).
• Applied pre-processing steps including bounding box encapsulation, dimension normalization, prostate region cropping, and image format conversion.
• Employed a pre-trained Faster Region Convolutional Neural Network (R-CNN) for seed detection.
• Evaluated model performance using leave-one-out cross-validation (LOOCV).

Main Results:

• The deep learning model achieved high performance metrics across all cases.
• Mean Average Precision (mAP) exceeded 0.91 in almost all cases.
• Mean Average Recall (mAR) was above 0.9 for 83.3% of cases.
• F1-scores consistently surpassed 0.91, with averaged results of 0.979 for mAP, 0.937 for mAR, and 0.957 for F1-score.

Conclusions:

• The developed deep learning model demonstrates high accuracy in automatically detecting implanted seeds on fluoroscopy images.
• Despite limitations in interpreting overlapping seeds, the model shows significant potential for enhancing prostate brachytherapy procedures.
• Further research and development could refine the model for broader clinical adoption.