Photon-counting detector of the ophthalmic artery: impact of virtual monoenergetic levels, reconstruction kernel, and quantum iterative reconstruction (QIR) on image quality.

Background: The ophthalmic artery (OA), a major branch of the internal carotid artery that supplies orbital vital structures and can serve as a potential biomarker for ocular and systemic diseases, is challenging to visualize due to its small diameter, tortuous course, and complex anatomy. Photon-counting detector computed tomography (PCD-CT) outperforms conventional energy-integrating detector computed tomography (CT) in orbital vascular imaging; however, relevant key reconstruction parameters for OA visualization have yet to be investigated. This study aimed to identify the optimal PCD-CT reconstruction parameters, including virtual monoenergetic imaging (VMI) level, reconstruction kernel, and quantum iterative reconstruction (QIR), for maximizing OA image quality. Methods: A total of 17 patients (33 OAs) underwent ultra-high-resolution carotid PCD-CT angiography. Axial images were reconstructed at 40-120 keV (20 keV increments), using reconstruction kernels [body vascular convolution (Bv) 40, 48, 56, 64, and 72] and QIR levels 3 and 4. For objective analysis, two blinded radiologists placed standardized regions of interest (ROIs) in the OA, intraorbital fat, and medial rectus muscle belly, respectively. Mean arterial signal intensity attenuation (MA), signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR) were measured. Two blinded radiologists subjectively assessed overall image quality, vessel sharpness, and diagnostic assess-ability (using 5-point Likert scales). Linear mixed-effects models (LMMs) identified optimal parameter combinations via a composite z-score. Friedman and Dunn's tests were used to compare the top protocols. Results: The objective analysis revealed that the MA, SNR, and CNR showed an inverse relationship with keV. The highest CNR was obtained at 40 keV using Bv40 and QIR4. Lower keV levels (40-60 keV) increased the MA by 136-195%, but decreased the SNR by 54.9-67.5% compared with higher keV levels (100-120 keV). The MA values initially increased with higher Bv values, peaking at 491.53±371.75 Hounsfield unit (HU) for Bv64. Conversely, both the SNR and CNR demonstrated a consistent decreasing trend with increasing Bv values. Sharper kernels (Bv72) decreased the CNR by 56.9% but increased the MA by 54.3% compared with smoother kernels (Bv40). QIR4 exhibited higher SNR and CNR values compared to those reconstructed with QIR3 (P<0.05). The MA values were similar between QIR3 and QIR4. QIR4 enhanced the SNR by 36.5%, the CNR by 37.6%, and the MA by 0.94% compared with QIR3. The subjective analysis showed that 40 keV/Bv56/QIR4 achieved perfect median scores (5/5) across all subjective features, significantly outperforming lower reconstruction kernel protocols (Bv40: median 1-2, Bv48: median 3-4; P<0.0001). Despite a superior objective CNR at Bv40, subjective ratings favored the Bv56 balanced noise-contrast profile. Subjective consensus identified 40 keV, Bv56 kernel, and QIR4 as optimal. Conclusions: For PCD-CT visualization of the OA, a VMI level of 40 keV combined with the Bv56 kernel and QIR4 provides the optimal balance between objective metrics and subjective radiologist preference, enhancing vascular delineation while controlling image noise. This protocol may improve diagnostic confidence in assessing small-caliber orbital vasculature.