Synthetic CT from biplanar radiographs for patellofemoral anatomical assessment: development and validation of an adapted X2CT-GAN model.
Authors
Affiliations (5)
Affiliations (5)
- Department of Orthopedic Biomechanical Laboratory, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China.
- School of Computer Science and Technology, East China Normal University, Shanghai, China.
- School of Computer Science and Technology, East China Normal University, Shanghai, China. [email protected].
- Department of Orthopedic Biomechanical Laboratory, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China. [email protected].
- Department of Orthopedics, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China. [email protected].
Abstract
Accurate assessment of osseous morphology is essential for the diagnosis and surgical management of patellofemoral instability. Methods capable of providing CT-like anatomical information while reducing radiation exposure are therefore of considerable clinical interest. This study aimed to validate the accuracy of radiography-derived synthetic computed tomography (sCT) generated by a deep learning model for anatomical assessment of patellofemoral instability. A total of 210 paired radiographs and CT scans were retrospectively collected and randomly divided into a training set (n = 170) and a validation set (n = 40). A deep learning model based on the X2CT-GAN framework was used to reconstruct sCT volumes from anteroposterior and lateral radiographs. Two blinded readers independently measured tibial tubercle-trochlear groove (TT-TG) distance, trochlear sulcus angle, and Caton-Deschamps index on both sCT and real CT images. Reliability, inter-modality agreement, and threshold-based classification agreement were evaluated. All intra- and inter-observer reliability values were excellent. Inter-modality agreement was excellent for TT-TG distance and good for trochlear sulcus angle. Agreement for patellar height was good. The mean absolute error for TT-TG distance ranged from 1.24 to 1.38 mm. Agreement for threshold-based classifications ranged from substantial to almost perfect, with high specificity observed across all evaluated cutoff values. Radiography-derived sCT demonstrated good agreement with real CT for both quantitative measurements and threshold-based classifications relevant to clinical decision-making in patellofemoral instability. These findings suggest that radiography-derived synthetic CT may represent a promising lower-radiation alternative for anatomical assessment of patellofemoral instability in selected clinical scenarios.