This investigation provided both quantitative and qualitative assessments of the Q.Clear reconstruction algorithm, contrasting a range of penalization factors (β value) with the prevalent OSEM + PSF algorithm (dubbed OSEM). Analyses were executed on NEMA phantoms filled with varying LBRs and on patients with rectal cancer administered with an 18F-FDG radiopharmaceutical.

The adeptness of Q.Clear to curb potential noise surges facilitates the pragmatic use of heightened iterations (about 25 in Q.Clear against 4 in OSEM). Q.Clear reconstructions exhibited superior \(}_}\), stemming from an increased iteration count and enhanced convergence. High noise in lower β values compromises image quality despite their precise quantification. Conversely, enhanced lesion visibility and consistent cold background were achieved with larger β values (Fig. 3). The BV for a β value of 300 is 5.8 ± 0.4, closely aligned with the BV of OSEM of 5.4 ± 0.4. The outcomes with NEMA phantom are similar to previous findings using NEMA, oval, and anthropomorphic phantom where a β value of 350 parallels OSEM in noise levels [3]. However, this contrasts subtly with Bjöersdorff et al.'s findings, who noted that OSEM reconstruction exhibited noise levels akin to a β of 550 at 1.5-min frames and β at 2.0-min frames [36].

Comparing OSEM and Q.Clear at matched noise levels (β = 300), Q.Clear-constructed images exhibited quantitative superiority over OSEM. For instance, measurements in a 22 mm sphere at an LBR of 4:1 indicated decreases of 21.7% in Δ\(}_-300)}\%\), Δ \(}_-300\right),}\) and 5.2% in Δ\(}_}}_-300)}\). Similar trends appeared in our clinical datasets. Caribé et al. substantiated that Q.Clear outperforms OSEM in tumor\(}_}\), \(}_}\), and contrast, even when the noise levels are equivalent [17].

The variation in BV resulting from reconstruction with different β values did not exhibit significant differences across various LBRs. The relative difference of BV demonstrated a difference of < 4.1% between LBRs of 2:1 and 8:1 and 3.9% between LBRs of 4:1 and 8:1. This was associated with an insignificant variation between background concentrations in the three LBR examinations (Fig. 2). Reynes-Llompart et al.'s [37] study, which utilized a NEMA phantom with LBRs 2:1, 4:1, and 8:1, reported a 1–2% variance in BV among three LBRs. However, we identified distinct trends for contrast, CR, and \(}_}\), where this relative difference more pronounced in lower LBRs (p < 0.05). A prior study indicated that, in the case of cold lesions in the phantom, the CNR also rises with an increase in the β value, even if the LBR does not seem to influence it [37].

In Q.Clear, relative differences in \(}_}\), CR and contrast were more significant for smaller lesions, attributed to the incorporation of PSF modeling into this algorithm. \(\Delta }_\)% dropped by 155.2, 128.8, and 68.1 for the smallest sphere (10 mm); by 26.8, 15.8, and 10.9 for the mid-sized (22 mm); and by 10.3, 7.5, and 4.8 for the largest sphere (37 mm) when LBRs of 2:1, 4:1, and 8:1 were considered, respectively. Our clinical study mirrored these observations. The escalation in quantitative parameters in \(_\) for small lesions was at least double that of large lesions (Fig. 6). For instance, the relative difference in SNR, SBR, contrast, and \(}_}\) between OSEM and \(.\mathrm}_\) decreased by 57.7%, 13.6%, 12.9%, and 12.0% for small lesions, and by 38.9%, 4.2%, 3.4%, and 3.6% for large lesions.

In our phantom study with an LBR of 8:1, all Q.Clear reconstructions yielded significantly lower LE than OSEM. Thus, \(\Delta }_-500}\mathrm\) increased by 54.5 (p < 0.05, Table 2). Our findings align with those of Elin Lindström. They found that LE was elevated for OSEM compared to β values of 133, 267, 400, and 533 [26].

Both phantom and clinical data indicated that quantitative parameters shifted rapidly as the β value escalated to 300. Nevertheless, minor alterations at high β values were reported. \(\Delta }_}}_\mathrm\), \(\Delta }_\mathrm\) and \(\Delta }_\mathrm\) dwindled by up to 49.5, 46.8 and 37.1. Furthermore, \(\Delta }_}}_\mathrm\), \(\Delta }_\mathrm\) and \(\Delta }_\mathrm\) also receded by up to 16.0, 19.3, and 18.7 (Figs. 2, 4). Reynes-Llompart et al. [37] highlighted the relative difference of quantitative parameters in Q.Clear, when using a β value of > 500, plateaued.

Our results demonstrate a greater relative difference in CR and BV when using Q.Clear with a BGO scanner compared to the results of previous studies with an LYSO scanner. At an LBR 4:1, the \(\Delta }_\)% in spheres with 10 mm and 22 mm diameters decreased by 128.8 and 15.8, respectively. Teoh's study filled the NEMA phantom with a 4:1 ratio and scanned it using an LYSO scanner. Their findings indicated that the \(\Delta }_\)% in sphere with diameters of 10 mm and 22 mm was 100.2 and 8.4, respectively [14]. Our clinical investigation found that Q.Clear reconstruction yields greater improvement in quantitative parameters on BGO scanners compared to other scanners. Lesion \(}_}\), SNR and SBR increased by 18.1%, 52.3%, and 16.9%, respectively, in \(.\mathrm}_\) compared to OSEM. Lindström et al. conducted a clinical study using images obtained by an LSO scanner, showing that \(.\mathrm}_\), in comparison with OSEM, resulted in 11%, 22%, and 12% increases in \(}_}\), SNR and SBR, respectively[26].

Using a low penalizing parameter might increase noise, potentially leading to false positive enhancement when estimating lesion uptake or mistakenly identifying noise as lesions. Conversely, excessive smoothing in Q.Clear with higher β values could lead to reduced RC and, consequently, incorrect negative interpretations. Based on our phantom study, the optimal β value for small lesions ranges from 200 for LBR 2:1 to 300 for LBR 8:1, improving SUV while maintaining an acceptable noise level. For larger lesions, the optimal β value lies between 400 for LBR 2:1 and 500 for LBR 8:1 to enhance SNR. It is vital to recognize the variability in lesion ratios and the lack of preset lesion sizes within clinical studies. Detecting large lesions with high activity is relatively straightforward while pinpointing smaller lesions with low activity is more challenging. Therefore, if only one reconstruction is performed for diagnosis, a β value of 300 is recommended to achieve the most accurate interpretation of the images.

Numerous factors strongly influence the optimization of the Q.Clear algorithm in PET scanning. One pivotal factor is the application or purpose of the PET scan. Different clinical or research needs might necessitate varying levels of image resolution, noise suppression, or contrast enhancement. Contrast is another important element. Lesions with higher contrast (i.e., the pronounced difference in signal intensity compared to surrounding tissues) may require distinct optimization strategies than those with lower contrast. Images with different noise and varying lesion sizes demand unique optimization. The patient count in our study was below the desired number, and the limited range of lesion sizes constrained our analysis. Larger studies are needed to confirm our findings. Assessing the effects of Body Mass Index (BMI), scan duration, various PET applications, and injected dose on optimizing the β-factor for a broader patient population was not the focus of this investigation.

In conclusion, Q.Clear offers enhanced quantitative measurements while maintaining a noise level comparable to the OSEM algorithm. This results in superior image quality and lesion detection. The Q.Clear optimization depends on both lesion size and LBR. As lesion size and LBR decrease, the optimal β value follows suit. For rectal cancer cases, we suggest using \(.\mathrm}_\) for smaller lesions and \(.\mathrm}_\) for larger ones.