Deep Learning–Based Reconstruction in Ultra-Low-Dose CT of the Ankle and Foot: A Comparative Study with Hybrid Iterative Reconstruction

Abstract

Objectives: To evaluate the quantitative and qualitative performance of ultra-low-dose CT with deep learning image reconstruction (DLR) compared to conventional hybrid iterative reconstruction (IR) in patients with ankle and foot fractures. Methods: A total of 32 patients (mean age, 54 ± 16 years) with ankle and foot fractures were included in this retrospective study. All patients underwent ultra-low-dose CT imaging (effective dose, 0.86 ± 0.11 μSv), and the same CT raw data were reconstructed using both DLR and hybrid IR. Image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were quantitatively measured, and image sharpness was evaluated using a no-reference perceptual sharpness metric. Subjective image quality was assessed by a board-certified radiologist and an orthopedic surgeon using a five-point scale. Results: DLR significantly reduced image noise (47.33 ± 6.60 HU) compared to hybrid IR (87.65 ± 12.48 HU), and significantly improved SNR (31.42 ± 5.81 for DLR vs. 19.15 ± 4.48 for hybrid IR) and CNR (50.59 ± 7.62 for DLR vs. 24.08 ± 3.81 for hybrid IR). Image sharpness was significantly (p = 0.001) improved with DLR (0.76 ± 0.09) compared to hybrid IR (0.59 ± 0.09). Subjective image analysis revealed enhanced visualization of trabecular architecture, superior delineation and integrity of cortical bone, and improved depiction of bony cortical lesions with DLR compared to hybrid IR. Conclusions: Ultra-low-dose CT with DLR offers a superior approach to enhance image quality, making it a valuable tool for clinical practice.