Skeleton Segmentation on Bone Scintigraphy for BSI Computation.

Area of Science:

• Medical Imaging
• Oncology
• Computer Vision

Background:

• The Bone Scan Index (BSI) is a key imaging biomarker for quantifying bone metastasis in cancer patients.
• Accurate segmentation of both bones and metastatic lesions (hotspots) is essential for BSI calculation.
• Existing research primarily focuses on binary classification, with limited studies addressing pixel-wise bone segmentation.

Purpose of the Study:

• To compare the performance of three advanced Convolutional Neural Network (CNN) models for bone segmentation in scintigraphy.
• To evaluate the efficacy of these models on an in-house dataset for clinical application.

Main Methods:

• Three state-of-the-art CNN models were implemented and compared for bone segmentation.
• The models were evaluated using an in-house dataset comprising bone scintigraphy images.
• Performance was assessed using precision, sensitivity, and F1-score metrics via 10-fold cross-validation.

Main Results:

• Mask R-CNN achieved the highest performance among the evaluated models.
• For prostate cancer patients, Mask R-CNN obtained precision, sensitivity, and F1-scores of 0.93, 0.87, and 0.90, respectively.
• For breast cancer patients, Mask R-CNN achieved precision, sensitivity, and F1-scores of 0.92, 0.86, and 0.88, respectively.

Conclusions:

• Mask R-CNN is a highly effective model for pixel-wise bone segmentation in bone scintigraphy.
• The robust performance of Mask R-CNN indicates its reliability for clinical use in BSI calculation and cancer metastasis assessment.