Deep learning cascade networks for segmentation of fluorine-18 sodium fluoride positron emission tomography scans of equine metacarpo- and metatarsophalangeal joints outperform atlas-based method.

CNN-based segmentation of equine fetlock 18F-NaF PET scans outperforms atlas methods in speed and accuracy.

This study developed and validated a convolutional neural network (CNN)-based deep learning cascade system for automated segmentation of fluorine-18 sodium fluoride (18F-NaF) PET scans of equine fetlock joints. Researchers retrospectively collected 84 PET and CT scans from two facilities, manually labeling anatomical structures including the third metacarpal bone, proximal phalanx, proximal sesamoid bones, and soft tissue. These labeled datasets were used to train a cascade of CNNs with and without data augmentation. Performance was assessed using the Dice coefficient and compared against an established atlas-based segmentation method. On the test set of 8 scans, the CNN achieved mean Dice coefficients of 0.88 overall, outperforming the atlas-based approach (mean Dice 0.80). The CNN performed best segmenting the third metacarpal bone (Dmean = 0.92) and least accurately for the medial proximal sesamoid bone (Dmean = 0.82), though both exceeded atlas-based values. The CNN method also demonstrated superior processing speed. Accurate automated segmentation of fetlock PET images enables more precise quantification of radiotracer uptake in specific anatomical regions, facilitating improved lesion characterization and potentially enabling novel diagnostic strategies for detecting early or subclinical musculoskeletal pathology. This advancement is particularly relevant for performance horse medicine, where early identification of fetlock pathology—a common site of career-ending injuries—is critical. The study demonstrates that deep learning segmentation is a viable and superior alternative to atlas-based methods for equine nuclear medicine imaging workflows.

This summary was distilled by AI and may occasionally misinterpret data. Confirm critical details with the primary literature before clinical application.