Efficacy Study of Deep Progressive Reconstruction Algorithm in Enhancing 18F-FDG PET Image Quality for Breast Cancer Patients with Different Body Mass Index

Efficacy Study of Deep Progressive Reconstruction Algorithm in Enhancing 18F-FDG PET Image Quality for Breast Cancer Patients with Different Body Mass Index | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Efficacy Study of Deep Progressive Reconstruction Algorithm in Enhancing 18 F-FDG PET Image Quality for Breast Cancer Patients with Different Body Mass Index Qi An, Fei Wen, Min Zhao, Lingchao Li, Ningning Lv, Bin Chen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9025681/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Objective To evaluate the efficacy of Deep Progressive Reconstruction (DPR) versus Ordered Subset Expectation Maximization (OSEM) algorithms in enhancing ^18F-FDG PET image quality across different body mass index (BMI) strata in patients with breast cancer. Methods This retrospective study included patients with breast cancer who underwent diagnostic ^18F-FDG PET/CT at the China-Japan Union Hospital of Jilin University between June and December 2023. Whole-body PET/CT was performed using the United Imaging uMI 780 system, with images reconstructed using three algorithms: DPR, OSEM2 (two iterations), and OSEM3 (three iterations). Patients were stratified into four BMI subgroups: underweight, normal weight, overweight, and obese. Two board-certified nuclear radiologists independently and blindly assessed image quality, noise level, and lesion conspicuity using a 5-point Likert scale. Quantitative metrics included maximal lesion diameter, maximum standardized uptake value (SUVmax), peak SUV (SUVpeak), liver signal-to-noise ratio (LSNR), tumor-to-background ratio (T/N), and liver SUV standard deviation (SUVSD). Statistical analyses were performed with one-way analysis of variance (ANOVA) and Mann–Whitney U tests to compare the three reconstruction methods across BMI subgroups and lesion sizes. Results A total of 118 female patients with 188 measurable lesions were analyzed. DPR-reconstructed PET images achieved significantly higher quality scores and lower noise levels than OSEM2 and OSEM3 (all P < 0.05). Rising BMI significantly worsened image quality and noise scores in OSEM2/OSEM3 (P < 0.05), with OSEM3 in obese patients failing to meet diagnostic criteria (quality score: 2.6; noise score: 2.7). DPR reconstruction yielded significantly higher SUVmax, SUVpeak, LSNR, and T/N values, while producing markedly lower SUVSD compared to OSEM methods (all P < 0.05). Compared with OSEM3, DPR conferred greater improvements in SUVmax, SUVpeak, and T/N in low-weight patients (P = 0.011, 0.048, and 0.014, respectively), and larger gains in SUVSD and LSNR in high-weight patients (P = 0.003 and < 0.001, respectively). Inverse correlations were observed between lesion size and DPR improvements versus OSEM2 for SUVmax (r=–0.298), SUVpeak (r=–0.243), and T/N (r=–0.326) (all P < 0.01), and versus OSEM3 for SUVmax (r=–0.213, P < 0.01). When compared with OSEM2, DPR provided greater benefits for sub-2 cm lesions (+ 12% SUVmax, + 6% SUVpeak, + 17% T/N) than for larger lesions (+ 5%, + 1%, + 6%; all P < 0.001). Conclusion DPR significantly improved ^18F-FDG PET image quality, noise suppression, and lesion conspicuity compared with OSEM. These advantages were most pronounced in patients with higher BMI and in small lesions, highlighting the potential of DPR to enhance diagnostic performance in challenging patient subgroups. PET/CT Deep progressive reconstruction Image quality Body mass index Breast cancer Figures Figure 1 Figure 2 Introduction Globally, breast cancer incidence reached 2.3 million new cases in 2022, accounting for 25% of all female malignancies and representing the highest cancer incidence among women. Projections indicate a 38% rise in global cases by 2050, with the most rapid growth expected in low- and middle-income countries [ 1 ] . Evidence shows that obesity is significantly correlated with breast cancer development, progression, and clinical outcomes. Due to alterations in hormone levels, sustained chronic inflammation, and insulin resistance, obese women exhibit a 20–40% higher risk of breast cancer compared with normal-weight women, an association that is especially pronounced in postmenopausal patients [ 2 , 3 ] . With its superior sensitivity and positive predictive value, 18 F-fluorodeoxyglucose positron emission tomography/computed tomography ( 18 F-FDG PET/CT) has become a cornerstone technique for breast cancer diagnosis and staging [ 4 , 5 ] . However, PET/CT application in obese individuals presents notable challenges. Elevated body fat composition and increased soft-tissue thickness lead to extended photon trajectories and greater respiratory excursion amplitudes, predisposing PET imaging to substantial photon attenuation and scattering. Conventional ordered subset expectation maximization (OSEM) reconstruction algorithms often demonstrate limited efficacy in correcting these phenomena, resulting in reconstructed images with prominent motion artifacts and elevated noise, which degrade image contrast and spatial resolution. This degradation reduces diagnostic accuracy and increases the risk of underdetection or misclassification of small lesions [ 2 , 6 – 9 ] . Routine approaches such as augmenting radiopharmaceutical doses or extending PET acquisition times can moderately enhance the signal-to-noise ratio (SNR) in obese patients. Yet these measures substantially increase radiation burden while diminishing patient tolerance and comfort. Moreover, prolonged supine immobilization elevates the likelihood of severe motion artifacts from respiratory and body movements, further compromising diagnostic precision. Consequently, moving beyond traditional physical correction methods toward sophisticated reconstruction algorithms has emerged as a pivotal research priority, particularly for overweight and obese breast cancer patients [ 10 – 14 ] . Enabled by advances in artificial intelligence, the deep progressive reconstruction (DPR) algorithm integrates convolutional neural networks (CNNs) into the conventional OSEM framework, providing substantial noise suppression and markedly improved lesion conspicuity in PET images [ 15 ] . Validation studies have shown that DPR preserves diagnostic image quality while reducing the administered radiopharmaceutical activity to about one-third of conventional doses [ 16 ] . However, earlier investigations of DPR included heterogeneous tumor types, where outcomes were confounded by distinct tumor biology, variable acquisition protocols, and differing susceptibility to respiratory motion—factors limiting comparability and reproducibility of imaging biomarkers [ 16 , 17 ] . To address these limitations, we focused exclusively on breast cancer, a tumor entity with relatively uniform biological characteristics that requires high-fidelity PET monitoring. Using standardized acquisition and reconstruction protocols, we performed comprehensive, multidimensional comparisons of radiomic features across stratified body weight subgroups. Materials and methods Patients’ selection This retrospective study enrolled 118 patients with breast cancer who underwent 18 F-FDG PET/CT at the China-Japan Union Hospital of Jilin University between June 5 and December 13, 2023. The inclusion criteria were as follows: (i) Presence of PET/CT-positive lesions and (ii) surgical intervention within 2 weeks post-imaging with histopathological confirmation of breast cancer via paraffin-embedded tissue examination. The exclusion criteria were as follows: (i) evidence of liver metastases or preexisting liver pathology, (ii) concurrent additional malignancies, (iii) inaccessible raw imaging data, and (iv) history of invasive diagnostic procedures at the lesion site before PET/CT acquisition. The study protocol was approved by the Institutional Review Board of the China-Japan Union Hospital of Jilin University (Approval number: 2020042601) and adhered to the principles of the Declaration of Helsinki and national regulatory standards. Based on the WHO classification, patients were categorized by body mass index (BMI) into: Underweight (<18.5 kg/m²), Normal-weight (18.5–24.9 kg/m²), Overweight (25–29.9 kg/m²), and Obese (≥30 kg/m²). For analytical purposes, underweight and normal-weight patients were grouped into the low-body weight group, whereas overweight and obese patients formed the high-body-weight group [18] . Image acquisition and reconstruction 18 F-FDG was synthesized by the Department of Nuclear Medicine, China-Japan Union Hospital of Jilin University. 18 F− was produced via a medical cyclotron (HM-10, Sumitomo Heavy Industries, Tokyo, Japan) and subsequently incorporated into FDG using an automated synthesis module (PET-FDG-IT-NA, PET CO.,LTD. Beijing, China) with consistent radiochemical purity >95%, as validated by thin-layer chromatography. Patients were required to fast for at least 6 h and discontinue insulin therapy before imaging. Blood glucose levels were confirmed to be less than 10 mmol/L. Weight-adjusted 18 F-FDG (0.37 MBq/kg) was administered intravenously through the antecubital vein, followed by a 45-60 minute uptake period under resting conditions in a dimly lit, quiet room prior to PET/CT acquisition. Imaging was performed using a digital PET/CT scanner (uMI 780, United Imaging Healthcare, Shanghai, China) with a sensitivity of 16 kcps/MBq, a spatial resolution of 2.9 mm at 1 cm from the isocenter, and a time-of-flight (TOF) resolution of 450 ps. A low-dose CT scan was initially acquired for attenuation correction and anatomical localization using a fixed tube voltage (120 kV) and automatic tube current modulation (15-100 mA). The step-and-shoot scanning protocol covered the skull vertex to the mid-femur, with a 2.5-minute acquisition per bed position. Based on the patient height, 4-6 bed positions were obtained with a 35% overlap between adjacent beds to ensure seamless volumetric reconstruction. The PET data were reconstructed using both OSEM and DPR algorithms. OSEM reconstructions employed two configurations: two iterations (OSEM2) and three iterations (OSEM3) with 20 subsets each, Gaussian post-filtering (3 mm full width at half maximum), 192×192 matrix, 600 mm field of view (FOV), and 2.68 mm slice thickness, with TOF and point spread function (PSF) modeling corrections activated. All reconstructions incorporated standard corrections, including scatter, random coincidence, dead time, radioactive decay, CT-based attenuation, and detector normalization. DPR reconstructions maintained an identical matrix size (192×192), FOV (600 mm), and slice thickness (2.68 mm) as OSEM without the application of additional post-processing techniques. Image analysis All PET/CT images were independently interpreted by two board-certified nuclear medicine physicians, each with more than 5 years of clinical experience. A subjective assessment was conducted using a 5-point Likert scale to evaluate three key parameters: overall image quality, noise level, and lesion conspicuity [19] . Discrepancies in interpretation were adjudicated by a third senior nuclear medicine physician (≥10 years of experience), with final determinations requiring concordance from at least two evaluators. The subjective scoring system was defined as follows: 1 = non-diagnostic (severe noise, blurred lesions, and clinically unusable). 2 = Suboptimal (prominent noise/artefacts and faintly visible lesions with low diagnostic confidence). 3 = Acceptable (moderate artifacts/noise, lesions sufficiently visible for clinical diagnosis). 4 = Good (minimal artifacts/noise, clearly delineated lesions, and diagnostic reliability). 5 = Excellent (near-zero noise, exquisitely defined lesions, and absence of artifacts). Images scoring ≥3 were deemed diagnostically acceptable, while scores ≤2 were classified as non-diagnostic. A semi-quantitative analysis was conducted by an independent nuclear medicine physician using the uWI-MI workstation, with measurements focused on the liver parenchyma, aortic arch, and target lesions. Three 20-mm-diameter circular ROIs were placed in the homogeneous right liver lobe segments (excluding lesions/vascular structures) for SUV mean and Liver standard deviation (SUV SD ) quantification. The aortic arch SUV mean was measured using a single 20-mm circular ROI. Spherical 3D VOIs encompassing all pathologically verified lesions were generated to extract the SUV max and SUV peak . A maximum of six target lesions per patient were analyzed, prioritizing the three highest and three lowest FDG-avid lesions when >6 were present. Lesions were classified according to the longest axial diameter on CT as small (≤2 cm) or large (>2 cm). Manual image registration was applied when the PET/CT misalignment exceeded acceptable thresholds [20] . SUV SD was defined as the image noise. Liver signal-to-noise ratio (LSNR) = Liver SUV mean / SUV SD . Tumor-to-background ratio (T/N) = lesion SUV max / aortic arch SUV mean . The variation rate of SUV max (ΔSUV max ) = (SUV max from DPR reconstruction - SUV max from OSEM2 or OSEM3 reconstruction) / SUV max from OSEM2 or OSEM3 reconstruction. The variation rates of SUV peak , SUV SD , LSNR, and T/N (ΔSUV peak , ΔSUV SD , ΔLSNR, and ΔT/N) were calculated using the same method as ΔSUV max [20-22] . Statistical analysis All statistical analyses were conducted using SPSS Statistics software (version 27.0; IBM Corporation, Armonk, NY, USA). Quantitative variables conforming to a normal distribution are presented as mean ± standard deviation (SD). Non-normally distributed quantitative variables are reported as medians with interquartile ranges (IQR). Inter-rater agreement for subjective visual assessments was evaluated using weighted kappa (κ) statistics. For normally distributed variables, intergroup differences in quantitative parameters were compared using one-way analysis of variance (ANOVA). Non-normally distributed variables were analyzed using the Wilcoxon signed-rank test for paired comparisons. Inter-group differences in variation rates (ΔSUV max , ΔSUV peak , ΔSUV SD , ΔLSNR, ΔT/N) were assessed using Mann-Whitney U tests. Statistical significance was set at P < 0.05 all tests. Results Patient characteristics A total of 118 female patients with 188 pathologically confirmed breast lesions were enrolled in this study. The mean age was 54.3 ± 10.2 years (range: 29–85 years), and the mean body mass index (BMI) was 24.6 ± 4.3 kg/m² (range: 17.8–38.1 kg/m²). The mean maximum lesion diameter was 20.7 ± 12.5 mm (range: 4.4–96.0 mm). Postoperative pathology revealed invasive carcinoma of no special type (n = 99), invasive carcinoma of a special type (n = 7), ductal carcinoma in situ (n = 7), lobular carcinoma in situ (n = 4), and Paget’s disease (n = 1). Lesion distribution analysis showed unifocal disease in 83 patients (70.3%) and multifocal disease (≥2 lesions) in 35 patients (29.7%). By size stratification, 117 lesions (62.2%) were small (≤20 mm) and 71 lesions (37.8%) were large (>20 mm). Comparative analysis across BMI quartiles demonstrated no statistically significant differences in age (P = 0.086), maximum lesion diameter (P = 0.360), fasting blood glucose level (P = 0.251), or ultrasound BI-RADS category distribution (P = 0.409) (Table 1). Table 1 . Patient clinical characteristics (n=118) Parameters Underweight (n=10) Normal (n=62) Overweight (n=28) Obese (n=18) p Age (years) 50.50±6.91 52.95±10.58 55.29±10.99 59.28±10.29 0.086 BMI (kg/m 2 ) 18.20±0.23 22.46±1.67 26.82±1.21 31.74±2.40 <0.001** Maximum lesion diameter (mm) 15.12±7.55 21.17±13.45 20.46±12.53 21.89±11.04 0.360 Injection dosage (mCI) 4.90±0.39 5.90±0.54 6.83±0.58 8.10±1.24 <0.001** Fasting blood glucose levels (mmol/L) 5.59±0.55 5.89±1.05 6.25±1.00 5.89±0.92 0.251 Ultrasound BI-RADS 0.409 4a 1 1 0 0 4b 0 6 2 5 4c 6 31 14 9 5 3 24 12 4 Data are means±standard deviations, *P<0.05, **P<0.01 Subjective assessment Inter-reader agreement for subjective image quality evaluation demonstrated excellent concordance (weighted κ = 0.872, 95% CI: 0.837–0.896). DPR reconstruction achieved significantly higher ratings for both overall image quality and noise suppression compared with OSEM2 (P < 0.05), while OSEM2 performed significantly better than OSEM3 (P < 0.05). Lesion conspicuity scores did not differ significantly between DPR and OSEM3, although DPR was statistically superior to OSEM2. Progressive BMI elevation correlated with significant deterioration in image quality and noise scores for both OSEM2 and OSEM3, whereas DPR maintained relative stability with only minimal decline. In overweight and obese subgroups, DPR-generated images achieved quality scores comparable to those of normal-weight subjects reconstructed with OSEM2 and OSEM3, demonstrating BMI-robust performance. Diagnostic acceptability analysis showed that DPR and OSEM2 achieved 100% acceptable rates across all BMI categories. In contrast, OSEM3 yielded acceptable images in underweight, normal-weight, and overweight subgroups but failed in obese patients, producing non-diagnostic images (quality score: 2.6 ± 0.3; noise score: 2.7 ± 0.4). Comprehensive statistical outcomes are presented in Table 2. Table 2. Subjective visual scores among different BMI groups under various reconstruction protocols (n=118) Parameters DPR OSEM2 OSEM3 Overall image quality Underweight 5.0±0.0 5.0±0.0 5.0±0.0 Normal 5.0±0.1 4.7±0.6 4.2±0.8 Overweight 4.8±0.4 4.2±0.8 3.6±0.9 Obese 4.7±0.6 3.4±0.8 2.6±1.0 All 4.9±0.3 4.4±0.8 3.9±1.1 Noise Underweight 5.0±0.0 5.0±0.0 4.9±0.3 Normal 5.0±0.0 4.7±0.6 4.2±0.8 Overweight 4.9±0.3 4.2±0.8 3.6±0.9 Obese 4.7±0.6 3.2±0.9 2.7±1.3 All 4.9±0.3 4.4±0.9 3.9±1.1 Lesion conspicuity Underweight 5.0±0.0 4.1±0.8 4.6±0.5 Normal 4.9±0.5 4.5±1.0 4.8±0.6 Overweight 4.8±0.6 4.0±1.2 4.4±0.8 Obese 4.9±0.4 4.2±1.0 4.6±0.7 All 4.9±0.5 4.3±1.1 4.6±0.7 Data are means±standard deviations. Semi-quantitative analysis Semi-quantitative analysis demonstrated significantly elevated values in DPR reconstructed images versus OSEM2/OSEM3 for: SUV max , SUV peak , LSNR, and T/N. Conversely, SUV SD was reduced by 44.4% in DPR versus OSEM2 and by 55.6% versus OSEM3, indicating superior noise suppression (P<0.01). Crucially, the liver SUV mean showed no statistically significant variations across the reconstruction methods, confirming preserved metabolic quantification reliability (Table 3). Figure 1 shows the differences in the visual appearance of the three reconstruction algorithms on the axial PET images of the liver of a 56-year-old female patient. Table 3. Quantitative parameters in PET images reconstructed using different algorithms Parameters DPR OSEM2 OSEM3 SUV max 7.98 (4.90-11.94) 7.36 (4.27-10.77)** 7.52 (4.76-10.88)** SUV peak 5.39 (3.12-7.91) 4.99 (2.94-7.66)** 5.07 (3.04-7.66)** Live SUV mean 2.48 (2.30-2.76) 2.49 (2.26-2.79) 2.48 (2.28-2.80) SUV SD 0.20 (0.16-0.24) 0.36 (0.28-0.47)** 0.45 (0.34-0.53)** LSNR 13.00 (10.66-14.42) 7.14 (5.38-8.70)** 5.65 (4.65-6.82)** T/N 4.89 (3.06-7.75) 4.13 (2.68-7.08)** 4.27 (2.75-7.31)** Data are medians (range), *P<0.05, **P<0.01 As presented in Table 4, in comparison between the DPR and OSEM2 reconstruction groups, ΔLSNR was 12.5% greater in the higher-weight group than in the lower-weight group, showing a statistically significant difference (P=0.025). No statistically significant differences were observed in ΔSUV max (P=0.107), ΔSUV peak (P=0.128), ΔSUV SD (P=0.061), or T/N (P=0.149) between the weight groups. In the comparison between the DPR and OSEM3 reconstruction groups, the higher-weight group exhibited significantly higher LSNR (+14.0%, P<0.001) and ΔSUV SD (+5.6%, P=0.003) than the lower-weight group, whereas ΔSUV max (-28.6%, P=0.011), ΔSUV peak (-33.3%, P=0.048), and T/N (-38.5%, P=0.014) were significantly lower than those in the lower-weight group. Table 4. Quantitative differences in high- and low-body-weight groups across reconstruction schemes Parameters low-body-weight (n=72) high-body-weight (n=46) P DPR-OSEM2 △SUV max 0.10 (0.02，0.19) 0.07 (-0.01，0.15) 0.107 △SUV peak 0.04 (0.00，0.10) 0.02 (-0.02，0.08) 0.128 △SUV SD -0.42 (-0.47，-0.36) -0.43 (-0.53，-0.37) 0.061 △LSNR 0.72 (0.58，0.88) 0.81 (0.62，1.07) 0.025* △T/N 0.14 (0.03，0.27) 0.10 (-0.00，0.25) 0.149 DPR-OSEM3 △SUV max 0.07 (0.01，0.14) 0.05 (-0.04，0.09) 0.011* △SUV peak 0.03 (-0.01，0.07) 0.02 (-0.04，0.05) 0.048* △SUV SD -0.53 (-0.57，-0.48) -0.56 (-0.59，-0.53) 0.003** △LSNR 1.14 (0.98，1.35) 1.30 (1.17，1.42) <0.001** △T/N 0.13 (0.02，0.23) 0.08 (-0.05，0.17) 0.014* Data are medians (range), *P<0.05, **P<0.01 In comparisons between DPR and OSEM2 reconstruction groups, lesion size was negatively correlated with ΔSUV max (r=-0.298, 95% CI: 0.01 – 0.18, P< 0.01), ΔSUV peak (r=-0.243, 95% CI: -0.01 – 0.09, P< 0.01), and ΔT/N (r=-0.326, 95% CI: 0.01 – 0.26, P< 0.01). In the DPR versus OSEM3 reconstruction comparison, a significant negative correlation was observed solely between lesion size and ΔSUV max (r=-0.213, 95% CI: 0.01 – 0.12, P 0.05) or ΔT/N (r=-0.137, 95% CI: 0.02 – 0.21, P> 0.05) (Table 5). Relative to OSEM2, DPR significantly improved SUV max , SUV peak , and T/N in small lesions by 12%, 6%, and 17%, respectively, with significantly larger improvements than those in large lesions (5%, 1%, and 6%, respectively; all P < 0.001). For DPR versus OSEM3, small lesions exhibited a 9% increase in SUV max , significantly surpassing the improvement in large lesions (P < 0.001), whereas enhancements in SUV peak and T/N did not differ significantly according to lesion size (Table 6). Figure 2 presents the maximum intensity projection (MIP) images from four breast cancer patients with small lesions and differing BMI reconstructed using the OSEM2, OSEM3, and DPR algorithms. Table 5. Correlation between different lesion sizes and △SUV max , △SUV peak , △T/N Parameters DPR vs. OSEM2 R DPR vs. OSEM3 R △SUV max 0.09 (0.01，0.18) -0.298** 0.06 (-0.01，0.12) -0.213** △SUV peak 0.04 (-0.01，0.09) -0.243** 0.03 (-0.02，0.06) -0.051 △T/N 0.11 (0.01，0.26) -0.326** 0.09 (-0.02，0.21) -0.137 Data are medians (range), *P<0.05, **P<0.01 Table 6. Differences in △SUV max , △SUV peak , and △T/N among lesions of different sizes Parameters small lesions ( 2cm ) P DPR vs. OSEM2 △SUVmax 0.12 (0.04，0.25) 0.05 (0.01，0.09) <0.001** △SUVpeak 0.06 (0.00，0.12) 0.01 (-0.02，0.04) <0.001** △T/N 0.17 (0.06，0.31) 0.06 (-0.01，0.16) <0.001** DPR vs. OSEM3 △SUVmax 0.09 (0.00，0.16) 0.03 (-0.02，0.06) <0.001** △SUVpeak 0.04 (0.00， 0.08) 0.03 (-0.01，0.07) 0.065 △T/N 0.13 (0.02， 0.31) 0.10 (0.01，0.22) 0.082 Data are medians (range), *P<0.05, **P<0.01 Discussion The DPR technique represents a deep learning–based algorithm for PET image reconstruction that refines conventional OSEM through a progressive learning architecture. Evidence indicates that DPR achieves diagnostic image quality comparable to standard-dose OSEM using only one-third of the dose or scan time, thereby reducing radiation exposure and improving workflow efficiency [16] . Consistent with previous studies [17] , our results demonstrated that underweight patients achieved the highest image quality and noise suppression across all reconstruction methods. However, increasing BMI was associated with progressive deterioration in PET image quality and noise scores, although the decline was significantly attenuated with DPR compared with OSEM2 and OSEM3. Notably, OSEM3 reconstructions fell below diagnostic thresholds (quality: 2.6; noise: 2.7) in obese patients, whereas DPR maintained diagnostic acceptability. Importantly, DPR achieved image quality scores in obese patients comparable to those of OSEM2 and OSEM3 in normal-weight individuals, highlighting its robustness across BMI strata. The absence of significant inter-BMI differences in lesion conspicuity scores may reflect radiologists’ reliance on absolute lesion metabolic activity rather than relative contrast, allowing biological signal to outweigh algorithmic differences. In clinical practice, OSEM is limited by incomplete convergence due to progressive noise amplification with each iteration, necessitating a trade-off between reconstruction depth and noise suppression. This constraint results in partial convergence, which adversely affects both image quality and SUV quantification precision [23, 24] . In contrast, DPR-reconstructed images in our study showed significantly higher SUVmax, SUVpeak, LSNR, and T/N, along with markedly reduced SUVSD, compared with OSEM2 and OSEM3, consistent with prior observations by Hirji H et al. [25] . These improvements can be attributed to DPR’s integrated CNN framework, which enables multidimensional data optimization during iterative reconstruction. Furthermore, DPR incorporates physical effect modeling into its architecture, providing superior correction of scatter coincidence events and minimizing quantitative deviations in homogeneous tissues such as the liver [15] . By accurately modeling the point spread function, DPR also mitigates partial volume effects, resulting in substantial improvements in SUVmax, SUVpeak, and T/N. Similar findings were reported by Lv et al., who confirmed DPR’s benefits in noise suppression and contrast enhancement [15] . Nevertheless, DPR-driven SUVmax elevations present challenges for cross-method comparability. In longitudinal studies where baseline scans are reconstructed with OSEM and follow-up scans with DPR, discrepancies in SUV quantification may confound interpretation of ΔSUVmax as a biomarker of treatment response. Additionally, diagnostic thresholds validated for OSEM may not directly apply to DPR images. These concerns underscore the need for reconstruction-invariant quantitative frameworks and updated response assessment guidelines. Establishing standardized protocols for handling mixed-reconstruction datasets will be essential for the widespread clinical adoption of DPR. Our study revealed that the DPR algorithm exhibited a distinct reconstruction performance compared with OSEM3 across BMI-stratified patients. DPR reconstruction yielded significantly higher increases in SUV max , SUV peak , and T/N for lesions in patients compared than in high-BMI patients (P<0.05). Conversely, patients with high BMI demonstrated superior improvement in LSNR and a more pronounced reduction in SUV SD . This differential effect stems from DPR's adaptive optimization capability of the DPR; its progressive deep learning architecture employs multidimensional feature networks to dynamically balance noise suppression and feature enhancement. High-BMI patients prioritize noise reduction to substantially improve image homogeneity. In patients with low BMI, the algorithm enhances lesion contrast on high-quality baseline data, thereby significantly improving lesion conspicuity. Such BMI-dependent optimization differentials underscore DPR's intelligent capacity of DPR to adapt reconstruction strategies according to the intrinsic data quality characteristics. Conventional studies attribute partial volume effects (PVE) as the dominant factor compromising PET quantification accuracy for sub-centimeter lesions. Although OSEM reconstruction enhances the spatial resolution through iterations, its capacity to recover true metabolic activity in lesions <2 cm remains constrained [26, 27] , a limitation stemming from the progressive noise amplification inherent in the iterative process. Evidence confirms that the diagnostic sensitivity for sub-2 cm lesions is consistently inferior to that for larger lesions, with significantly higher false-negative rates [26, 28] . Our data demonstrated that DPR induced significant negative correlations between lesion diameter and improvements in SUV max , SUV peak , and T/N versus OSEM2 (P<0.01). This suggests DPR's improved efficacy of DPR in boosting small-lesion conspicuity, leveraging its distinctive multiscale feature extraction and hardware integration. Diverging from the OSEM signal recovery limitations caused by noise amplification, DPR processes data through multiscale structural inputs. This enables targeted feature extraction across spatial scales, effectively preserving true signals while suppressing background noise, thereby revealing metabolic information traditionally obscured by small lesions. Quantitatively, DPR elevated the SUV max by 12% (vs. OSEM2), SUV peak by 6%, and T/N by 17% for sub-2 cm lesions. Notably, it achieved a 9% higher SUV max than OSEM3, which is known for its superior lesion conspicuity compared with conventional algorithms. Visual assessment (Fig. 2) confirmed that DPR-reconstructed images exhibited sharper lesion margins and superior lesion-to-background contrast compared to OSEM reconstructions. This enhancement mechanism is derived from DPR's dual-action approach: suppressing image noise while amplifying edge information via convolutional neural networks and strategically weighting edge features during image fusion. Collectively, DPR enhances T/N through synergistic background noise reduction and lesion SUV max elevation, substantially improving both image contrast and lesion detectability. This advantage is most pronounced when mitigating OSEM's limitations of OSEM for small lesions. Fundamentally, while OSEM struggles to balance noise amplification and signal recovery during iterations, DPR achieves precise equilibrium via multiscale feature extraction and hardware-accelerated processing. Consequently, it demonstrated superior small-lesion detectability and quantitative accuracy, which are critical advancements in early cancer detection. This study has several limitations. First, the single-center design and restricted sample size necessitate future large-scale multicenter validations. Critical priorities include verifying the stability and reproducibility of the DPR-driven SUV max quantification across heterogeneous PET scanner platforms. Second, while focusing on quantitative imaging metrics, this study did not evaluate clinical endpoints such as diagnostic efficacy or therapy response prediction. Future studies should directly compare DPR with conventional algorithms in terms of clinically critical outcomes, including lesion detection sensitivity and diagnostic accuracy. Conclusion Compared with conventional OSEM, DPR significantly improved the overall image quality and noise suppression in 18 F-FDG PET for breast cancer patients, effectively ameliorating image degradation caused by increased tissue attenuation in obese patients (BMI ≥25 kg/m²). Notably, DPR showed significantly better improvement in key quantitative metrics for small lesions (<2 cm in diameter) than for large lesions, which was negatively correlated with lesion size. These findings highlight DPR's unique value of DPR in optimizing breast tumor PET imaging, particularly for obese patients, and for small lesion detection. Declarations Ethics approval and consent to participate Ethics approval was obtained from the research Ethics Committee of China-Japan Union Hospital of JiLin University. This retrospectively study was carried out in the accordance with Declaration of Helsinkiand, and informed consent was waived due to its retrospective nature. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Funding No funding was applied for none was received. Author Contribution B.C. conceptualized and designed the study; Q. A. collected the data and drafted the initial manuscript; F.W. and M. Z. performed the statistical analysis and interpreted the results; L.L.C. prepared Figures; N.N.L. critically revised the manuscript for important intellectual content. Data Availability All data supporting the findings of this study are available within the article and its Supplementary Information files. References Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229-63. Fan L, Strasser-Weippl K, Li JJ, St Louis J, Finkelstein DM, Yu KD, et al. Breast cancer in China. Lancet Oncol. 2014;15(7):e279-89. Clinton SK, Giovannucci EL, Hursting SD. The World Cancer Research Fund/American Institute for Cancer Research Third Expert Report on Diet, Nutrition, Physical Activity, and Cancer: Impact and Future Directions. J Nutr. 2020;150(4):663-671. Ko H, Baghdadi Y, Love C, Sparano JA. Clinical Utility of 18F-FDG PET/CT in Staging Localized Breast Cancer Before Initiating Preoperative Systemic Therapy. J Natl Compr Canc Netw. 2020;18(9):1240-6. Han S, Choi JY. Impact of 18F-FDG PET, PET/CT, and PET/MRI on Staging and Management as an Initial Staging Modality in Breast Cancer: A Systematic Review and Meta-analysis. Clin Nucl Med. 2021;46(4):271-82. Dwivedi P, Sawant V, Vajarkar V, Vatsa R, Choudhury S, Jha AK, et al. Analysis of image quality by regulating beta function of BSREM reconstruction algorithm and comparison with conventional reconstructions in carcinoma breast studies of PET CT with BGO detector. Nucl Med Commun. 2023;44(1):56-64. Rubello D, Colletti PM. SUV Harmonization Between Different Hybrid PET/CT Systems. Clin Nucl Med. 2018;43(11):811-4. Xiao J, Yu H, Sui X, Hu Y, Cao Y, Liu G, et al. Can the BMI-based dose regimen be used to reduce injection activity and to obtain a constant image quality in oncological patients by 18F-FDG total-body PET/CT imaging? Eur J Nucl Med Mol Imaging. 2021;49(1):269-78. Tahari AK, Chien D, Azadi JR, Wahl RL. Optimum lean body formulation for correction of standardized uptake value in PET imaging. J Nucl Med. 2014;55(9):1481-4. Matsubara K, Ibaraki M, Nemoto M, Watabe H, Kimura Y. A review on AI in PET imaging. Ann Nucl Med. 2022;36(2):133-43. Tsuchiya J, Yokoyama K, Yamagiwa K, Watanabe R, Kimura K, Kishino M, et al. Deep learning-based image quality improvement of 18F-fluorodeoxyglucose positron emission tomography: a retrospective observational study. EJNMMI Phys. 2021;8(1):31. Usmani S, Marafi F, Ahmed N, Esmail A, Al Kandari F, Van den Wyngaert T. Diagnostic Challenge of Staging Metastatic Bone Disease in the Morbidly Obese Patients: A Primary Study Evaluating the Usefulness of 18F-Sodium Fluoride (NaF) PET-CT. Clin Nucl Med. 2017;42(11):829-36. Sanaat A, Shiri I, Arabi H, Mainta I, Nkoulou R, Zaidi H. Deep learning-assisted ultra-fast/low-dose whole-body PET/CT imaging. Eur J Nucl Med Mol Imaging. 2021;48(8):2405-15. Xu L, Yang R, Li RS, Liu RC, Meng QL, Wang F. Investigate the quantification accuracy of small lesions in oncological 18F-FDG PET/CT using a deep progressive learning reconstruction method. BMC Med Imaging. 2026;26(1):84. Lv Y, Xi C. PET image reconstruction with deep progressive learning. Phys Med Biol. 2021;66(10). Wang T, Qiao W, Wang Y, Wang J, Lv Y, Dong Y, et al. Deep progressive learning achieves whole-body low-dose 18F-FDG PET imaging. EJNMMI Phys. 2022;9(1):82. Yang H, Chen S, Qi M, Chen W, Kong Q, Zhang J, et al. Investigation of PET image quality with acquisition time/bed and enhancement of lesion quantification accuracy through deep progressive learning. EJNMMI Phys. 2024;11(1):7. Obesity: preventing and managing the global epidemic. Report of a WHO consultation. World Health Organ Tech Rep Ser. 2000;894:i-xii, 1-253. Sonni I, Baratto L, Park S, Hatami N, Srinivas S, Davidzon G, et al. Initial experience with a SiPM-based PET/CT scanner: influence of acquisition time on image quality. EJNMMI Phys. 2018;5(1):9. Adams MC, Turkington TG, Wilson JM, Wong TZ. A systematic review of the factors affecting accuracy of SUV measurements. AJR Am J Roentgenol. 2010;195(2):310-20. Messerli M, Stolzmann P, Egger-Sigg M, Trinckauf J, D'Aguanno S, Burger IA, et al. Impact of a Bayesian penalized likelihood reconstruction algorithm on image quality in novel digital PET/CT: clinical implications for the assessment of lung tumors. EJNMMI Phys. 2018;5(1):27. McDermott GM, Chowdhury FU, Scarsbrook AF. Evaluation of noise equivalent count parameters as indicators of adult whole-body FDG-PET image quality. Ann Nucl Med. 2013;27(9):855-61. Ahn S, Ross SG, Asma E, Miao J, Jin X, Cheng L, et al. Quantitative comparison of OSEM and penalized likelihood image reconstruction using relative difference penalties for clinical PET. Phys Med Biol. 2015;60(15):5733-51. Zan K, Duan Y, Zhao M, Li H, Cui X, Chai L, et al. Performance of the Iterative OSEM and HYPER Algorithm for Total-body PET at SUVmax with a Low 18F-FDG Activity, a Short Acquisition Time and Small Lesions. Curr Med Imaging. 2024;20:e15734056274225.. Hirji H, Sullivan K, Lasker I, Sharif MS, Nunes A, Shepherd C, et al. Effect of PET Image Reconstruction Techniques on Unexpected Aorta Uptake. Mol Imaging Radionucl Ther. 2019;28(1):1-7. Iwano S, Ito S, Tsuchiya K, Kato K, Naganawa S. What causes false-negative PET findings for solid-type lung cancer? Lung Cancer. 2013;79(2):132-6. Adler S, Seidel J, Choyke P, Knopp MV, Binzel K, Zhang J, et al. Minimum lesion detectability as a measure of PET system performance. EJNMMI Phys. 2017;4(1):13. Khalaf M, Abdel-Nabi H, Baker J, Shao Y, Lamonica D, Gona J. Relation between nodule size and 18F-FDG-PET SUV for malignant and benign pulmonary nodules. J Hematol Oncol. 2008;1:13. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 01 May, 2026 Reviewers agreed at journal 09 Apr, 2026 Reviewers invited by journal 03 Apr, 2026 Editor invited by journal 10 Mar, 2026 Editor assigned by journal 09 Mar, 2026 Submission checks completed at journal 09 Mar, 2026 First submitted to journal 03 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9025681","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":617101097,"identity":"efe3a120-500a-4b1e-be6b-85be9e43bd9e","order_by":0,"name":"Qi An","email":"","orcid":"","institution":"China-Japan Union Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"An","suffix":""},{"id":617101100,"identity":"c74a7c4b-13b2-41a6-b712-c531f79bda1f","order_by":1,"name":"Fei Wen","email":"","orcid":"","institution":"China-Japan Union Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Fei","middleName":"","lastName":"Wen","suffix":""},{"id":617101102,"identity":"3bdd8044-1abc-4075-89ac-0be8f66c1693","order_by":2,"name":"Min Zhao","email":"","orcid":"","institution":"China-Japan Union Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Min","middleName":"","lastName":"Zhao","suffix":""},{"id":617101114,"identity":"6205d7e0-e6f8-43fb-9afa-eb722092a0cc","order_by":3,"name":"Lingchao Li","email":"","orcid":"","institution":"China-Japan Union Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Lingchao","middleName":"","lastName":"Li","suffix":""},{"id":617101115,"identity":"6ef52f53-100d-4069-a151-3c0617d25713","order_by":4,"name":"Ningning Lv","email":"","orcid":"","institution":"China-Japan Union Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Ningning","middleName":"","lastName":"Lv","suffix":""},{"id":617101116,"identity":"313536ee-fa64-41da-b8d0-a9bdbcdf5dbb","order_by":5,"name":"Bin Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2klEQVRIie3RMQrCMBSA4ZSAXQpdXxHv8KAgBT1Mi2AXEccOgq8U4uha6DG8QELAqe4dHHqEHqCIrZNTzSiYfwmB9/EIYcxm+8XAIQkI3Geahis3ITxvo8PaDUgZE7fArNv6KB1D4le5gAb1PFRKAMtWCbl3Ob3koURQog6XciR1mpC3jycJQiLmgHrzJo7QCYGH30mP+nSlkTzNSIGAW45sJGRAhrfkLeCag1RFFN/SUHi7aeJX51ZCP3xlqVXTHVeLi1tPk499krF4OGeG8+M+Mp+12Wy2/+oFWXNJ3w4k/xQAAAAASUVORK5CYII=","orcid":"","institution":"China-Japan Union Hospital of Jilin University","correspondingAuthor":true,"prefix":"","firstName":"Bin","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2026-03-04 04:38:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9025681/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9025681/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106544459,"identity":"f702548a-6a8f-41b1-8e9a-326736c089fb","added_by":"auto","created_at":"2026-04-09 16:40:59","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":54041,"visible":true,"origin":"","legend":"\u003cp\u003eAxial PET images of the liver plane reconstructed by (A) OSEM2 (SUV\u003csub\u003eSD\u003c/sub\u003e = 0.36), (B) OSEM3 (SUV\u003csub\u003eSD\u003c/sub\u003e = 0.45), and (C) DPR (SUV\u003csub\u003eSD\u003c/sub\u003e = 0.19) in a 56-year-old female patient with breast cancer. Compared with OSEM2/3 images, the DPR-reconstructed image shows lower noise and a clearer, smoother appearance.\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9025681/v1/4039c2abbac5bd2f39802adf.jpg"},{"id":106544381,"identity":"b11aad29-1acc-480b-9129-d842e60f5095","added_by":"auto","created_at":"2026-04-09 16:40:40","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":245809,"visible":true,"origin":"","legend":"\u003cp\u003eMaximum Intensity Projection (MIP) images reconstructed by OSEM2, OSEM3, and DPR algorithms for four breast cancer patients with different BMIs. (A-C) The long diameter of the breast lesion in the underweight patient is 1.64 cm, with SUV\u003csub\u003emax\u003c/sub\u003e of 3.84, 4.22, and 4.77 respectively; (D-F) The long diameter of the breast lesion in the normal-weight patient is 1.32 cm, with SUV\u003csub\u003emax\u003c/sub\u003e of 3.31, 3.59, and 4.42 respectively; (G-I) The long diameter of the breast lesion in the overweight patient is 1.52 cm, with SUV\u003csub\u003emax\u003c/sub\u003e of 3.94, 4.26, and 5.23 respectively; (J-L) The long diameter of the breast lesion in the obese patient is 1.66 cm, with SUV\u003csub\u003emax\u003c/sub\u003e of 7.05, 7.50, and 9.07 respectively. Compared with OSEM2/3 images, DPR-reconstructed images show higher lesion conspicuity and lower noise.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9025681/v1/e9c19803da0159faff67a2db.jpg"},{"id":106544652,"identity":"a8527f35-0a29-4eec-8c98-b89b96d1c489","added_by":"auto","created_at":"2026-04-09 16:41:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1146600,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9025681/v1/29a6aa1f-a63b-49e2-9ffc-a094385a4a65.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eEfficacy Study of Deep Progressive Reconstruction Algorithm in Enhancing \u003csup\u003e18\u003c/sup\u003eF-FDG PET Image Quality for Breast Cancer Patients with Different Body Mass Index\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eGlobally, breast cancer incidence reached 2.3\u0026nbsp;million new cases in 2022, accounting for 25% of all female malignancies and representing the highest cancer incidence among women. Projections indicate a 38% rise in global cases by 2050, with the most rapid growth expected in low- and middle-income countries \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. Evidence shows that obesity is significantly correlated with breast cancer development, progression, and clinical outcomes. Due to alterations in hormone levels, sustained chronic inflammation, and insulin resistance, obese women exhibit a 20\u0026ndash;40% higher risk of breast cancer compared with normal-weight women, an association that is especially pronounced in postmenopausal patients \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eWith its superior sensitivity and positive predictive value, \u003csup\u003e18\u003c/sup\u003eF-fluorodeoxyglucose positron emission tomography/computed tomography (\u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT) has become a cornerstone technique for breast cancer diagnosis and staging \u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e. However, PET/CT application in obese individuals presents notable challenges. Elevated body fat composition and increased soft-tissue thickness lead to extended photon trajectories and greater respiratory excursion amplitudes, predisposing PET imaging to substantial photon attenuation and scattering. Conventional ordered subset expectation maximization (OSEM) reconstruction algorithms often demonstrate limited efficacy in correcting these phenomena, resulting in reconstructed images with prominent motion artifacts and elevated noise, which degrade image contrast and spatial resolution. This degradation reduces diagnostic accuracy and increases the risk of underdetection or misclassification of small lesions \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan additionalcitationids=\"CR7 CR8\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eRoutine approaches such as augmenting radiopharmaceutical doses or extending PET acquisition times can moderately enhance the signal-to-noise ratio (SNR) in obese patients. Yet these measures substantially increase radiation burden while diminishing patient tolerance and comfort. Moreover, prolonged supine immobilization elevates the likelihood of severe motion artifacts from respiratory and body movements, further compromising diagnostic precision. Consequently, moving beyond traditional physical correction methods toward sophisticated reconstruction algorithms has emerged as a pivotal research priority, particularly for overweight and obese breast cancer patients \u003csup\u003e[\u003cspan additionalcitationids=\"CR11 CR12 CR13\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eEnabled by advances in artificial intelligence, the deep progressive reconstruction (DPR) algorithm integrates convolutional neural networks (CNNs) into the conventional OSEM framework, providing substantial noise suppression and markedly improved lesion conspicuity in PET images \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. Validation studies have shown that DPR preserves diagnostic image quality while reducing the administered radiopharmaceutical activity to about one-third of conventional doses \u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. However, earlier investigations of DPR included heterogeneous tumor types, where outcomes were confounded by distinct tumor biology, variable acquisition protocols, and differing susceptibility to respiratory motion\u0026mdash;factors limiting comparability and reproducibility of imaging biomarkers \u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTo address these limitations, we focused exclusively on breast cancer, a tumor entity with relatively uniform biological characteristics that requires high-fidelity PET monitoring. Using standardized acquisition and reconstruction protocols, we performed comprehensive, multidimensional comparisons of radiomic features across stratified body weight subgroups.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003e\u003cstrong\u003ePatients\u0026rsquo; selection\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis retrospective study enrolled 118\u0026nbsp;patients with breast cancer\u0026nbsp;who underwent \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT at the China-Japan Union Hospital of Jilin University between June 5 and December 13, 2023. The inclusion criteria were\u0026nbsp;as follows: (i) Presence of PET/CT-positive lesions\u0026nbsp;and (ii) surgical intervention within 2 weeks post-imaging with histopathological confirmation of breast cancer via paraffin-embedded tissue examination.\u0026nbsp;The exclusion criteria were as follows: (i) evidence of\u0026nbsp;liver\u0026nbsp;metastases or preexisting liver pathology, (ii) concurrent additional malignancies, (iii) inaccessible raw imaging data,\u0026nbsp;and (iv)\u0026nbsp;history of invasive diagnostic procedures at the lesion site before PET/CT acquisition.\u0026nbsp;The study protocol was approved by the Institutional Review Board of\u0026nbsp;the China-Japan Union Hospital of Jilin University\u0026nbsp;\u0026nbsp;(Approval number: 2020042601)\u0026nbsp;and adhered to the principles of the Declaration of Helsinki and national regulatory standards.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBased on the WHO classification, patients were categorized by\u0026nbsp;body mass index\u0026nbsp;(BMI)\u0026nbsp;into: Underweight (\u0026lt;18.5 kg/m\u0026sup2;), Normal-weight (18.5\u0026ndash;24.9 kg/m\u0026sup2;), Overweight (25\u0026ndash;29.9 kg/m\u0026sup2;), and Obese (\u0026ge;30 kg/m\u0026sup2;).\u0026nbsp;For analytical purposes, underweight and normal-weight patients\u0026nbsp;were grouped into the low-body\u0026nbsp;weight\u0026nbsp;group, whereas overweight and obese patients formed the high-body-weight\u0026nbsp;group \u003csup\u003e[18]\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImage acquisition and reconstruction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e18\u003c/sup\u003eF-FDG was synthesized by the Department of Nuclear Medicine, China-Japan Union Hospital of Jilin University. \u003csup\u003e18\u003c/sup\u003eF\u0026minus; was produced via a medical cyclotron (HM-10, Sumitomo Heavy Industries, Tokyo, Japan) and subsequently incorporated into FDG using an automated synthesis module (PET-FDG-IT-NA, PET CO.,LTD. Beijing, China) with\u0026nbsp;consistent radiochemical purity\u0026nbsp;\u0026gt;95%, as validated by thin-layer chromatography. Patients were required to\u0026nbsp;fast for at least 6 h and discontinue insulin therapy before imaging. Blood glucose levels\u0026nbsp;were confirmed to be less than 10\u0026nbsp;mmol/L. Weight-adjusted \u003csup\u003e18\u003c/sup\u003eF-FDG (0.37 MBq/kg) was administered intravenously through the antecubital vein, followed by a 45-60 minute uptake period under resting conditions in a dimly lit, quiet room prior to PET/CT acquisition.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eImaging was performed using a digital PET/CT scanner (uMI 780, United Imaging Healthcare, Shanghai, China) with a sensitivity of 16 kcps/MBq,\u0026nbsp;a spatial resolution of 2.9 mm at 1 cm from the isocenter, and a time-of-flight (TOF) resolution of 450 ps.\u0026nbsp;A low-dose CT scan was initially acquired for attenuation correction and anatomical localization using\u0026nbsp;a fixed tube voltage (120 kV) and automatic tube current modulation (15-100 mA). The step-and-shoot scanning protocol covered\u0026nbsp;the skull vertex to\u0026nbsp;the mid-femur, with\u0026nbsp;a 2.5-minute acquisition per bed position. Based on\u0026nbsp;the patient height, 4-6 bed positions were obtained with a 35% overlap between adjacent beds to ensure seamless volumetric reconstruction.\u0026nbsp;The\u0026nbsp;PET data were reconstructed using both OSEM and DPR algorithms. OSEM reconstructions employed two configurations: two iterations (OSEM2) and three iterations (OSEM3) with 20 subsets each, Gaussian post-filtering (3 mm full width at half maximum), 192\u0026times;192 matrix, 600 mm field of view (FOV), and 2.68 mm slice thickness, with TOF and point spread function (PSF) modeling corrections activated.\u0026nbsp;All reconstructions incorporated standard corrections, including scatter, random coincidence, dead time, radioactive decay, CT-based attenuation, and detector normalization. DPR reconstructions maintained\u0026nbsp;an identical matrix size (192\u0026times;192), FOV (600 mm), and slice thickness (2.68 mm) as OSEM\u0026nbsp;without\u0026nbsp;the application of additional post-processing techniques.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImage analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll PET/CT images were independently interpreted by two board-certified nuclear medicine physicians, each with more than 5 years of clinical experience. A subjective assessment was conducted using a 5-point Likert scale\u0026nbsp;to evaluate three key parameters: overall image quality, noise level, and lesion conspicuity \u003csup\u003e[19]\u003c/sup\u003e.\u0026nbsp;Discrepancies in interpretation were adjudicated by a third senior nuclear medicine physician (\u0026ge;10 years of experience), with final determinations requiring concordance from at least two evaluators.\u0026nbsp;The subjective scoring system was defined as follows:\u0026nbsp;1 = non-diagnostic (severe noise, blurred lesions,\u0026nbsp;and clinically unusable).\u0026nbsp;2 = Suboptimal (prominent noise/artefacts and faintly visible lesions with low diagnostic confidence).\u0026nbsp;3 = Acceptable (moderate artifacts/noise, lesions sufficiently visible for clinical diagnosis).\u0026nbsp;4 = Good (minimal artifacts/noise, clearly delineated lesions,\u0026nbsp;and diagnostic\u0026nbsp;reliability).\u0026nbsp;5 = Excellent (near-zero noise, exquisitely defined lesions,\u0026nbsp;and absence of artifacts).\u0026nbsp;Images scoring \u0026ge;3 were deemed diagnostically acceptable, while scores \u0026le;2 were classified as non-diagnostic.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eA semi-quantitative analysis was conducted by an independent nuclear medicine physician using the uWI-MI workstation, with measurements focused on\u0026nbsp;the liver parenchyma, aortic arch, and target lesions.\u0026nbsp;Three 20-mm-diameter circular ROIs were placed in\u0026nbsp;the homogeneous right\u0026nbsp;liver\u0026nbsp;lobe segments (excluding lesions/vascular structures) for SUV\u003csub\u003emean\u003c/sub\u003e and\u0026nbsp;Liver\u0026nbsp;standard deviation (SUV\u003csub\u003eSD\u003c/sub\u003e) quantification.\u0026nbsp;The aortic arch SUV\u003csub\u003emean\u003c/sub\u003e was measured using a single 20-mm circular ROI.\u0026nbsp;Spherical 3D VOIs encompassing all pathologically verified lesions were generated to extract\u0026nbsp;the SUV\u003csub\u003emax\u003c/sub\u003e and SUV\u003csub\u003epeak\u003c/sub\u003e.\u0026nbsp;A maximum of six target lesions per patient were analyzed, prioritizing\u0026nbsp;the\u0026nbsp;three highest and three lowest FDG-avid lesions when \u0026gt;6 were present.\u0026nbsp;Lesions were classified according to\u0026nbsp;the\u0026nbsp;longest axial diameter on CT as small (\u0026le;2 cm) or large (\u0026gt;2 cm).\u0026nbsp;Manual image registration was applied when\u0026nbsp;the PET/CT misalignment exceeded acceptable thresholds\u0026nbsp;\u003csup\u003e[20]\u003c/sup\u003e.\u0026nbsp;SUV\u003csub\u003eSD\u003c/sub\u003e was defined as\u0026nbsp;the image noise. Liver signal-to-noise ratio (LSNR) =\u0026nbsp;Liver\u0026nbsp;SUV\u003csub\u003emean\u003c/sub\u003e / SUV\u003csub\u003eSD\u003c/sub\u003e. Tumor-to-background ratio (T/N) = lesion SUV\u003csub\u003emax\u003c/sub\u003e / aortic arch SUV\u003csub\u003emean\u003c/sub\u003e. The variation rate of SUV\u003csub\u003emax\u003c/sub\u003e (\u0026Delta;SUV\u003csub\u003emax\u003c/sub\u003e) = (SUV\u003csub\u003emax\u003c/sub\u003e from DPR reconstruction - SUV\u003csub\u003emax\u003c/sub\u003e from OSEM2 or OSEM3 reconstruction) / SUV\u003csub\u003emax\u003c/sub\u003e from OSEM2 or OSEM3 reconstruction. The variation rates of SUV\u003csub\u003epeak\u003c/sub\u003e, SUV\u003csub\u003eSD\u003c/sub\u003e, LSNR, and T/N (\u0026Delta;SUV\u003csub\u003epeak\u003c/sub\u003e, \u0026Delta;SUV\u003csub\u003eSD\u003c/sub\u003e, \u0026Delta;LSNR, and \u0026Delta;T/N) were calculated using the same method as \u0026Delta;SUV\u003csub\u003emax\u003c/sub\u003e \u003csup\u003e[20-22]\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll statistical analyses were conducted using SPSS Statistics software (version 27.0; IBM Corporation, Armonk, NY, USA). Quantitative variables conforming to\u0026nbsp;a normal distribution\u0026nbsp;are presented as mean \u0026plusmn; standard deviation (SD).\u0026nbsp;Non-normally distributed quantitative variables are reported as medians with interquartile ranges (IQR).\u0026nbsp;Inter-rater agreement for subjective visual assessments was evaluated using weighted kappa (\u0026kappa;) statistics.\u0026nbsp;For normally distributed variables, intergroup differences in quantitative parameters were compared using one-way analysis of variance (ANOVA).\u0026nbsp;Non-normally distributed variables were analyzed using\u0026nbsp;the Wilcoxon signed-rank test\u0026nbsp;for paired comparisons.\u0026nbsp;Inter-group differences in variation rates (\u0026Delta;SUV\u003csub\u003emax\u003c/sub\u003e, \u0026Delta;SUV\u003csub\u003epeak\u003c/sub\u003e, \u0026Delta;SUV\u003csub\u003eSD\u003c/sub\u003e, \u0026Delta;LSNR, \u0026Delta;T/N) were assessed using Mann-Whitney U tests. Statistical significance was set at P \u0026lt; 0.05 all tests.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003ePatient characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA total of 118 female patients with 188 pathologically confirmed breast lesions were enrolled in this study. The mean age was 54.3 \u0026plusmn; 10.2 years (range: 29\u0026ndash;85 years), and the mean body mass index (BMI) was 24.6 \u0026plusmn; 4.3 kg/m\u0026sup2; (range: 17.8\u0026ndash;38.1 kg/m\u0026sup2;). The mean maximum lesion diameter was 20.7 \u0026plusmn; 12.5 mm (range: 4.4\u0026ndash;96.0 mm). Postoperative pathology revealed invasive carcinoma of no special type (n = 99), invasive carcinoma of a special type (n = 7), ductal carcinoma in situ (n = 7), lobular carcinoma in situ (n = 4), and Paget\u0026rsquo;s disease (n = 1). Lesion distribution analysis showed unifocal disease in 83 patients (70.3%) and multifocal disease (\u0026ge;2 lesions) in 35 patients (29.7%). By size stratification, 117 lesions (62.2%) were small (\u0026le;20 mm) and 71 lesions (37.8%) were large (\u0026gt;20 mm). Comparative analysis across BMI quartiles demonstrated no statistically significant differences in age (P = 0.086), maximum lesion diameter (P = 0.360), fasting blood glucose level (P = 0.251), or ultrasound BI-RADS category distribution (P = 0.409) (Table 1).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e. Patient clinical characteristics (n=118)\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"580\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUnderweight\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n=10)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNormal\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n=62)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOverweight\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n=28)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eObese\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n=18)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ep\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003eAge (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e50.50\u0026plusmn;6.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e52.95\u0026plusmn;10.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e55.29\u0026plusmn;10.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e59.28\u0026plusmn;10.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e0.086\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003eBMI (kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e18.20\u0026plusmn;0.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e22.46\u0026plusmn;1.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e26.82\u0026plusmn;1.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e31.74\u0026plusmn;2.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e\u0026lt;0.001**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003eMaximum lesion diameter\u0026nbsp;(mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e15.12\u0026plusmn;7.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e21.17\u0026plusmn;13.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e20.46\u0026plusmn;12.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e21.89\u0026plusmn;11.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e0.360\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003eInjection dosage (mCI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e4.90\u0026plusmn;0.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e5.90\u0026plusmn;0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e6.83\u0026plusmn;0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e8.10\u0026plusmn;1.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e\u0026lt;0.001**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003eFasting blood glucose levels\u0026nbsp;(mmol/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e5.59\u0026plusmn;0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e5.89\u0026plusmn;1.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e6.25\u0026plusmn;1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e5.89\u0026plusmn;0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e0.251\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003eUltrasound BI-RADS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e0.409\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003e4a\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003e4b\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003e4c\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 34.715%;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5078%;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.9534%;\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8169%;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.2988%;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.7081%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eData are means\u0026plusmn;standard deviations, *P\u0026lt;0.05, **P\u0026lt;0.01\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSubjective assessment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInter-reader agreement for subjective image quality evaluation demonstrated excellent concordance (weighted \u0026kappa; = 0.872, 95% CI: 0.837\u0026ndash;0.896). DPR reconstruction achieved significantly higher ratings for both overall image quality and noise suppression compared with OSEM2 (P \u0026lt; 0.05), while OSEM2 performed significantly better than OSEM3 (P \u0026lt; 0.05). Lesion conspicuity scores did not differ significantly between DPR and OSEM3, although DPR was statistically superior to OSEM2. Progressive BMI elevation correlated with significant deterioration in image quality and noise scores for both OSEM2 and OSEM3, whereas DPR maintained relative stability with only minimal decline. In overweight and obese subgroups, DPR-generated images achieved quality scores comparable to those of normal-weight subjects reconstructed with OSEM2 and OSEM3, demonstrating BMI-robust performance. Diagnostic acceptability analysis showed that DPR and OSEM2 achieved 100% acceptable rates across all BMI categories. In contrast, OSEM3 yielded acceptable images in underweight, normal-weight, and overweight subgroups but failed in obese patients, producing non-diagnostic images (quality score: 2.6 \u0026plusmn; 0.3; noise score: 2.7 \u0026plusmn; 0.4). Comprehensive statistical outcomes are presented in Table 2.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u003c/strong\u003e Subjective visual scores among different BMI groups under various reconstruction protocols (n=118)\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDPR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOSEM2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOSEM3\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eOverall image quality\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eUnderweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5.0\u0026plusmn;0.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5.0\u0026plusmn;0.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5.0\u0026plusmn;0.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5.0\u0026plusmn;0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.7\u0026plusmn;0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.2\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eOverweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.8\u0026plusmn;0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.2\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e3.6\u0026plusmn;0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eObese\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e4.7\u0026plusmn;0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e3.4\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e2.6\u0026plusmn;1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eAll\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.9\u0026plusmn;0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.4\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e3.9\u0026plusmn;1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eNoise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eUnderweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5.0\u0026plusmn;0.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5.0\u0026plusmn;0.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.9\u0026plusmn;0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5.0\u0026plusmn;0.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.7\u0026plusmn;0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.2\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eOverweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.9\u0026plusmn;0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.2\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e3.6\u0026plusmn;0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eObese\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e4.7\u0026plusmn;0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e3.2\u0026plusmn;0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e2.7\u0026plusmn;1.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eAll\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.9\u0026plusmn;0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.4\u0026plusmn;0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e3.9\u0026plusmn;1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eLesion conspicuity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eUnderweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5.0\u0026plusmn;0.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.1\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.6\u0026plusmn;0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.9\u0026plusmn;0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.5\u0026plusmn;1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.8\u0026plusmn;0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eOverweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.8\u0026plusmn;0.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.0\u0026plusmn;1.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.4\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eObese\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e4.9\u0026plusmn;0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e4.2\u0026plusmn;1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003e4.6\u0026plusmn;0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 25%;\"\u003e\n \u003cp\u003eAll\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.9\u0026plusmn;0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.3\u0026plusmn;1.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4.6\u0026plusmn;0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eData are means\u0026plusmn;standard deviations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSemi-quantitative analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSemi-quantitative analysis demonstrated significantly elevated values in DPR reconstructed images versus OSEM2/OSEM3 for: SUV\u003csub\u003emax\u003c/sub\u003e, SUV\u003csub\u003epeak\u003c/sub\u003e, LSNR, and T/N. Conversely, SUV\u003csub\u003eSD\u003c/sub\u003e was reduced by 44.4% in DPR versus OSEM2 and by 55.6% versus OSEM3, indicating superior noise suppression (P\u0026lt;0.01). Crucially, the liver SUV\u003csub\u003emean\u003c/sub\u003e showed no statistically significant variations across the reconstruction methods, confirming preserved metabolic quantification reliability (Table 3). Figure 1 shows the differences in the visual appearance of the three reconstruction algorithms on the axial PET images of the liver of a 56-year-old female patient.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3.\u003c/strong\u003e Quantitative parameters in PET images reconstructed using different algorithms\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 16.4286%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDPR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOSEM2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOSEM3\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16.4286%;\"\u003e\n \u003cp\u003eSUV\u003csub\u003emax\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e7.98 (4.90-11.94)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e7.36 (4.27-10.77)**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e7.52 (4.76-10.88)**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16.4286%;\"\u003e\n \u003cp\u003eSUV\u003csub\u003epeak\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e5.39 (3.12-7.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e4.99 (2.94-7.66)**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e5.07 (3.04-7.66)**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16.4286%;\"\u003e\n \u003cp\u003eLive SUV\u003csub\u003emean\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e2.48 (2.30-2.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e2.49 (2.26-2.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e2.48 (2.28-2.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16.4286%;\"\u003e\n \u003cp\u003eSUV\u003csub\u003eSD\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e0.20 (0.16-0.24)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e0.36 (0.28-0.47)**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e0.45 (0.34-0.53)**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16.4286%;\"\u003e\n \u003cp\u003eLSNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e13.00 (10.66-14.42)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e7.14 (5.38-8.70)**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e5.65 (4.65-6.82)**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16.4286%;\"\u003e\n \u003cp\u003eT/N\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e4.89 (3.06-7.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e4.13 (2.68-7.08)**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 27.8571%;\"\u003e\n \u003cp\u003e4.27 (2.75-7.31)**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eData are medians (range), *P\u0026lt;0.05, **P\u0026lt;0.01\u003c/p\u003e\n\u003cp\u003eAs presented in Table 4, in comparison between the DPR and OSEM2 reconstruction groups, \u0026Delta;LSNR was 12.5% greater in the higher-weight group than in the lower-weight group, showing a statistically significant difference (P=0.025). No statistically significant differences were observed in \u0026Delta;SUV\u003csub\u003emax\u003c/sub\u003e (P=0.107), \u0026Delta;SUV\u003csub\u003epeak\u003c/sub\u003e (P=0.128), \u0026Delta;SUV\u003csub\u003eSD\u003c/sub\u003e (P=0.061), or T/N (P=0.149) between the weight groups. In the comparison between the DPR and OSEM3 reconstruction groups, the higher-weight group exhibited significantly higher LSNR (+14.0%, P\u0026lt;0.001) and \u0026Delta;SUV\u003csub\u003eSD\u003c/sub\u003e (+5.6%, P=0.003) than the lower-weight group, whereas \u0026Delta;SUV\u003csub\u003emax\u003c/sub\u003e (-28.6%, P=0.011), \u0026Delta;SUV\u003csub\u003epeak\u003c/sub\u003e (-33.3%, P=0.048), and T/N (-38.5%, P=0.014) were significantly lower than those in the lower-weight group.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4.\u003c/strong\u003e Quantitative differences in high- and low-body-weight groups across reconstruction schemes\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"552\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e\u003cstrong\u003elow-body-weight (n=72)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ehigh-body-weight (n=46)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003eDPR-OSEM2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△SUV\u003csub\u003emax\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.10 (0.02，0.19)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.07 (-0.01，0.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.107\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△SUV\u003csub\u003epeak\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.04 (0.00，0.10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.02 (-0.02，0.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.128\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△SUV\u003csub\u003eSD\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e-0.42 (-0.47，-0.36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e-0.43 (-0.53，-0.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.061\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△LSNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.72 (0.58，0.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.81 (0.62，1.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.025*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△T/N\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.14 (0.03，0.27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.10 (-0.00，0.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.149\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003eDPR-OSEM3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△SUV\u003csub\u003emax\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.07 (0.01，0.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.05 (-0.04，0.09)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.011*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△SUV\u003csub\u003epeak\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.03 (-0.01，0.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.02 (-0.04，0.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.048*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△SUV\u003csub\u003eSD\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e-0.53 (-0.57，-0.48)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e-0.56 (-0.59，-0.53)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.003**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△LSNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e1.14 (0.98，1.35)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e1.30 (1.17，1.42)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e\u0026lt;0.001**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 21.2341%;\"\u003e\n \u003cp\u003e△T/N\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.13 (0.02，0.23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32.6679%;\"\u003e\n \u003cp\u003e0.08 (-0.05，0.17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.4301%;\"\u003e\n \u003cp\u003e0.014*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eData are medians (range), *P\u0026lt;0.05, **P\u0026lt;0.01\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn comparisons between DPR and OSEM2 reconstruction groups, lesion size was negatively correlated with \u0026Delta;SUV\u003csub\u003emax\u003c/sub\u003e (r=-0.298, 95% CI: 0.01 \u0026ndash; 0.18, P\u0026lt; 0.01), \u0026Delta;SUV\u003csub\u003epeak\u003c/sub\u003e (r=-0.243, 95% CI: -0.01 \u0026ndash; 0.09, P\u0026lt; 0.01), and \u0026Delta;T/N (r=-0.326, 95% CI: 0.01 \u0026ndash; 0.26, P\u0026lt; 0.01). In the DPR versus OSEM3 reconstruction comparison, a significant negative correlation was observed solely between lesion size and \u0026Delta;SUV\u003csub\u003emax\u003c/sub\u003e (r=-0.213, 95% CI: 0.01 \u0026ndash; 0.12, P\u0026lt; 0.01), whereas no significant associations existed for \u0026Delta;SUV\u003csub\u003epeak\u003c/sub\u003e (r=-0.051, 95% CI: 0.02 \u0026ndash; 0.06, P\u0026gt; 0.05) or \u0026Delta;T/N (r=-0.137, 95% CI: 0.02 \u0026ndash; 0.21, P\u0026gt; 0.05) (Table 5).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRelative to OSEM2, DPR significantly improved SUV\u003csub\u003emax\u003c/sub\u003e, SUV\u003csub\u003epeak\u003c/sub\u003e, and T/N in small lesions by 12%, 6%, and 17%, respectively, with significantly larger improvements than those in large lesions (5%, 1%, and 6%, respectively; all P \u0026lt; 0.001). For DPR versus OSEM3, small lesions exhibited a 9% increase in SUV\u003csub\u003emax\u003c/sub\u003e, significantly surpassing the improvement in large lesions (P \u0026lt; 0.001), whereas enhancements in SUV\u003csub\u003epeak\u003c/sub\u003e and T/N did not differ significantly according to lesion size (Table 6). Figure 2 presents the maximum intensity projection (MIP) images from four breast cancer patients with small lesions and differing BMI reconstructed using the OSEM2, OSEM3, and DPR algorithms.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 5.\u003c/strong\u003e Correlation between different lesion sizes and △SUV\u003csub\u003emax\u003c/sub\u003e, △SUV\u003csub\u003epeak\u003c/sub\u003e, △T/N\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"549\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18.1818%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26.1818%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDPR vs. OSEM2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.7273%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 26.1818%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDPR vs. OSEM3\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.7273%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18.1818%;\"\u003e\n \u003cp\u003e△SUV\u003csub\u003emax\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1818%;\"\u003e\n \u003cp\u003e0.09 (0.01，0.18)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.7273%;\"\u003e\n \u003cp\u003e-0.298**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1818%;\"\u003e\n \u003cp\u003e0.06 (-0.01，0.12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.7273%;\"\u003e\n \u003cp\u003e-0.213**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18.1818%;\"\u003e\n \u003cp\u003e△SUV\u003csub\u003epeak\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1818%;\"\u003e\n \u003cp\u003e0.04 (-0.01，0.09)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.7273%;\"\u003e\n \u003cp\u003e-0.243**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1818%;\"\u003e\n \u003cp\u003e0.03 (-0.02，0.06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.7273%;\"\u003e\n \u003cp\u003e-0.051\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18.1818%;\"\u003e\n \u003cp\u003e△T/N\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1818%;\"\u003e\n \u003cp\u003e0.11 (0.01，0.26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.7273%;\"\u003e\n \u003cp\u003e-0.326**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1818%;\"\u003e\n \u003cp\u003e0.09 (-0.02，0.21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.7273%;\"\u003e\n \u003cp\u003e-0.137\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eData are medians (range), *P\u0026lt;0.05, **P\u0026lt;0.01\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 6.\u003c/strong\u003e Differences in △SUV\u003csub\u003emax\u003c/sub\u003e, △SUV\u003csub\u003epeak\u003c/sub\u003e, and △T/N among lesions of different sizes\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"553\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e\u003cstrong\u003esmall lesions\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;(\u0026lt;\u003c/strong\u003e\u003cstrong\u003e2cm)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e\u003cstrong\u003elarge lesions\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;(\u0026gt;\u003c/strong\u003e\u003cstrong\u003e2cm\u003c/strong\u003e\u003cstrong\u003e)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003eDPR vs. OSEM2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003e△SUVmax\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.12 (0.04，0.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.05 (0.01，0.09)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e\u0026lt;0.001**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003e△SUVpeak\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.06 (0.00，0.12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.01 (-0.02，0.04)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e\u0026lt;0.001**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003e△T/N\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.17 (0.06，0.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.06 (-0.01，0.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e\u0026lt;0.001**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003eDPR vs. OSEM3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003e△SUVmax\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.09 (0.00，0.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.03 (-0.02，0.06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e\u0026lt;0.001**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003e△SUVpeak\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.04 (0.00，\u0026nbsp;0.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.03 (-0.01，0.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e0.065\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 24.9548%;\"\u003e\n \u003cp\u003e△T/N\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.13 (0.02，\u0026nbsp;0.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 29.6564%;\"\u003e\n \u003cp\u003e0.10 (0.01，0.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15.7324%;\"\u003e\n \u003cp\u003e0.082\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eData are medians (range), *P\u0026lt;0.05, **P\u0026lt;0.01\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe DPR technique represents a deep learning\u0026ndash;based algorithm for PET image reconstruction that refines conventional OSEM through a progressive learning architecture. Evidence indicates that DPR achieves diagnostic image quality comparable to standard-dose OSEM using only one-third of the dose or scan time, thereby reducing radiation exposure and improving workflow efficiency\u003csup\u003e\u0026nbsp;[16]\u003c/sup\u003e. Consistent with previous studies \u003csup\u003e[17]\u003c/sup\u003e, our results demonstrated that underweight patients achieved the highest image quality and noise suppression across all reconstruction methods. However, increasing BMI was associated with progressive deterioration in PET image quality and noise scores, although the decline was significantly attenuated with DPR compared with OSEM2 and OSEM3. Notably, OSEM3 reconstructions fell below diagnostic thresholds (quality: 2.6; noise: 2.7) in obese patients, whereas DPR maintained diagnostic acceptability. Importantly, DPR achieved image quality scores in obese patients comparable to those of OSEM2 and OSEM3 in normal-weight individuals, highlighting its robustness across BMI strata. The absence of significant inter-BMI differences in lesion conspicuity scores may reflect radiologists\u0026rsquo; reliance on absolute lesion metabolic activity rather than relative contrast, allowing biological signal to outweigh algorithmic differences.\u003c/p\u003e\n\u003cp\u003eIn clinical practice, OSEM is limited by incomplete convergence due to progressive noise amplification with each iteration, necessitating a trade-off between reconstruction depth and noise suppression. This constraint results in partial convergence, which adversely affects both image quality and SUV quantification precision\u003csup\u003e[23, 24]\u003c/sup\u003e. In contrast, DPR-reconstructed images in our study showed significantly higher SUVmax, SUVpeak, LSNR, and T/N, along with markedly reduced SUVSD, compared with OSEM2 and OSEM3, consistent with prior observations by Hirji H et al. \u003csup\u003e[25]\u003c/sup\u003e. These improvements can be attributed to DPR\u0026rsquo;s integrated CNN framework, which enables multidimensional data optimization during iterative reconstruction. Furthermore, DPR incorporates physical effect modeling into its architecture, providing superior correction of scatter coincidence events and minimizing quantitative deviations in homogeneous tissues such as the liver\u0026nbsp;\u003csup\u003e[15]\u003c/sup\u003e.\u0026nbsp;By accurately modeling the point spread function, DPR also mitigates partial volume effects, resulting in substantial improvements in SUVmax, SUVpeak, and T/N. Similar findings were reported by Lv et al., who confirmed DPR\u0026rsquo;s benefits in noise suppression and contrast enhancement \u003csup\u003e[15]\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eNevertheless, DPR-driven SUVmax elevations present challenges for cross-method comparability. In longitudinal studies where baseline scans are reconstructed with OSEM and follow-up scans with DPR, discrepancies in SUV quantification may confound interpretation of \u0026Delta;SUVmax as a biomarker of treatment response. Additionally, diagnostic thresholds validated for OSEM may not directly apply to DPR images. These concerns underscore the need for reconstruction-invariant quantitative frameworks and updated response assessment guidelines. Establishing standardized protocols for handling mixed-reconstruction datasets will be essential for the widespread clinical adoption of DPR.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur study revealed that the DPR algorithm exhibited\u0026nbsp;a distinct reconstruction performance compared\u0026nbsp;with OSEM3 across BMI-stratified patients. DPR reconstruction yielded significantly higher increases in SUV\u003csub\u003emax\u003c/sub\u003e, SUV\u003csub\u003epeak\u003c/sub\u003e, and T/N for lesions in patients compared than in high-BMI patients (P\u0026lt;0.05). Conversely, patients\u0026nbsp;with high BMI demonstrated superior improvement in LSNR and a more pronounced reduction in SUV\u003csub\u003eSD\u003c/sub\u003e.\u0026nbsp;This differential effect stems from DPR\u0026apos;s adaptive optimization capability\u0026nbsp;of the DPR; its progressive deep learning architecture employs multidimensional feature networks to dynamically balance noise suppression and feature enhancement. High-BMI patients prioritize noise reduction to substantially improve image homogeneity.\u0026nbsp;In patients\u0026nbsp;with low BMI, the algorithm enhances lesion contrast on high-quality baseline data, thereby significantly improving lesion conspicuity.\u0026nbsp;Such BMI-dependent optimization differentials underscore DPR\u0026apos;s intelligent capacity\u0026nbsp;of DPR to adapt reconstruction strategies according to the intrinsic data quality characteristics.\u003c/p\u003e\n\u003cp\u003eConventional studies attribute partial volume effects (PVE) as the dominant factor compromising PET quantification accuracy for sub-centimeter lesions.\u0026nbsp;Although OSEM reconstruction enhances\u0026nbsp;the spatial resolution through iterations, its capacity to recover true metabolic activity in lesions \u0026lt;2 cm remains constrained\u0026nbsp;\u003csup\u003e[26, 27]\u003c/sup\u003e, a limitation stemming from\u0026nbsp;the progressive noise amplification inherent in the iterative process.\u0026nbsp;Evidence confirms that\u0026nbsp;the diagnostic sensitivity for sub-2 cm lesions is consistently inferior to that for larger lesions, with significantly higher false-negative rates\u0026nbsp;\u003csup\u003e[26, 28]\u003c/sup\u003e.\u0026nbsp;Our data demonstrated that DPR\u0026nbsp;induced significant negative correlations between lesion diameter and improvements in SUV\u003csub\u003emax\u003c/sub\u003e, SUV\u003csub\u003epeak\u003c/sub\u003e, and T/N versus OSEM2 (P\u0026lt;0.01). This suggests DPR\u0026apos;s improved efficacy of DPR in boosting small-lesion conspicuity, leveraging its distinctive multiscale feature extraction and hardware integration.\u0026nbsp;Diverging from\u0026nbsp;the OSEM\u0026nbsp;signal recovery limitations caused by noise amplification, DPR processes data through multiscale structural inputs. This enables targeted feature extraction across spatial scales, effectively preserving true signals while suppressing background noise, thereby revealing metabolic information traditionally obscured by small lesions.\u0026nbsp;Quantitatively, DPR elevated\u0026nbsp;the SUV\u003csub\u003emax\u003c/sub\u003e by 12% (vs. OSEM2), SUV\u003csub\u003epeak\u003c/sub\u003e by 6%, and T/N by 17% for sub-2\u0026nbsp;cm lesions. Notably, it achieved a 9% higher SUV\u003csub\u003emax\u003c/sub\u003e than OSEM3, which is known for its superior lesion conspicuity compared with conventional algorithms.\u0026nbsp;Visual assessment (Fig. 2) confirmed that DPR-reconstructed images exhibited sharper lesion margins and superior lesion-to-background contrast compared to OSEM reconstructions.\u0026nbsp;This enhancement mechanism\u0026nbsp;is derived from DPR\u0026apos;s dual-action approach: suppressing image noise while amplifying edge information via convolutional neural networks and strategically weighting edge features during image fusion.\u0026nbsp;Collectively, DPR enhances T/N through synergistic background noise reduction and lesion SUV\u003csub\u003emax\u003c/sub\u003e elevation, substantially improving both image contrast and lesion detectability. This advantage is most pronounced when mitigating OSEM\u0026apos;s limitations\u0026nbsp;of OSEM for small lesions. Fundamentally, while OSEM struggles to balance noise amplification and signal recovery during iterations, DPR achieves precise equilibrium via multiscale feature extraction and hardware-accelerated processing. Consequently, it demonstrated superior small-lesion detectability and quantitative accuracy, which are critical advancements in early cancer detection.\u003c/p\u003e\n\u003cp\u003eThis study has several limitations.\u0026nbsp;First, the single-center design and restricted sample size necessitate future large-scale multicenter validations. Critical priorities include verifying the stability and reproducibility of the DPR-driven SUV\u003csub\u003emax\u003c/sub\u003e quantification across heterogeneous PET scanner platforms. Second, while focusing on quantitative imaging metrics, this study did not evaluate clinical endpoints such as diagnostic efficacy or therapy response prediction. Future studies should directly compare DPR with conventional algorithms in terms of clinically critical outcomes, including lesion detection sensitivity and diagnostic accuracy.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eCompared with conventional OSEM, DPR significantly improved\u0026nbsp;the overall image quality and noise suppression in\u0026nbsp;\u003csup\u003e18\u003c/sup\u003eF-FDG PET for breast cancer patients, effectively ameliorating image degradation caused by increased tissue attenuation in obese patients (BMI \u0026ge;25 kg/m\u0026sup2;). Notably, DPR showed significantly better improvement in key quantitative metrics for small lesions (\u0026lt;2\u0026nbsp;cm in diameter)\u0026nbsp;than for large lesions,\u0026nbsp;which was negatively correlated with lesion size.\u0026nbsp;These findings highlight DPR\u0026apos;s unique value\u0026nbsp;of DPR in optimizing breast tumor PET imaging,\u0026nbsp;particularly for obese patients, and for small lesion detection.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEthics approval was obtained from the research Ethics Committee of China-Japan Union Hospital of JiLin University. This retrospectively study was carried out in the accordance with Declaration of Helsinkiand, and informed consent was waived due to its retrospective nature.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo funding was applied for none was received.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eB.C. conceptualized and designed the study; Q. A. collected the data and drafted the initial manuscript; F.W. and M. Z. performed the statistical analysis and interpreted the results; L.L.C. prepared Figures; N.N.L. critically revised the manuscript for important intellectual content.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data supporting the findings of this study are available within the article and its Supplementary Information files.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eBray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229-63.\u003c/li\u003e\n \u003cli\u003eFan L, Strasser-Weippl K, Li JJ, St Louis J, Finkelstein DM, Yu KD, et al. Breast cancer in China. Lancet Oncol. 2014;15(7):e279-89.\u003c/li\u003e\n \u003cli\u003eClinton SK, Giovannucci EL, Hursting SD. The World Cancer Research Fund/American Institute for Cancer Research Third Expert Report on Diet, Nutrition, Physical Activity, and Cancer: Impact and Future Directions. J Nutr. 2020;150(4):663-671.\u003c/li\u003e\n \u003cli\u003eKo H, Baghdadi Y, Love C, Sparano JA. Clinical Utility of 18F-FDG PET/CT in Staging Localized Breast Cancer Before Initiating Preoperative Systemic Therapy. J Natl Compr Canc Netw. 2020;18(9):1240-6.\u003c/li\u003e\n \u003cli\u003eHan S, Choi JY. Impact of 18F-FDG PET, PET/CT, and PET/MRI on Staging and Management as an Initial Staging Modality in Breast Cancer: A Systematic Review and Meta-analysis. Clin Nucl Med. 2021;46(4):271-82.\u003c/li\u003e\n \u003cli\u003eDwivedi P, Sawant V, Vajarkar V, Vatsa R, Choudhury S, Jha AK, et al. Analysis of image quality by regulating beta function of BSREM reconstruction algorithm and comparison with conventional reconstructions in carcinoma breast studies of PET CT with BGO detector. Nucl Med Commun. 2023;44(1):56-64.\u003c/li\u003e\n \u003cli\u003eRubello D, Colletti PM. SUV Harmonization Between Different Hybrid PET/CT Systems. Clin Nucl Med. 2018;43(11):811-4.\u003c/li\u003e\n \u003cli\u003eXiao J, Yu H, Sui X, Hu Y, Cao Y, Liu G, et al. Can the BMI-based dose regimen be used to reduce injection activity and to obtain a constant image quality in oncological patients by\u0026nbsp;18F-FDG total-body PET/CT imaging? Eur J Nucl Med Mol Imaging. 2021;49(1):269-78.\u003c/li\u003e\n \u003cli\u003eTahari AK, Chien D, Azadi JR, Wahl RL. Optimum lean body formulation for correction of standardized uptake value in PET imaging. J Nucl Med. 2014;55(9):1481-4.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eMatsubara K, Ibaraki M, Nemoto M, Watabe H, Kimura Y. A review on AI in PET imaging. Ann Nucl Med. 2022;36(2):133-43.\u003c/li\u003e\n \u003cli\u003eTsuchiya J, Yokoyama K, Yamagiwa K, Watanabe R, Kimura K, Kishino M, et al. Deep learning-based image quality improvement of\u0026nbsp;18F-fluorodeoxyglucose positron emission tomography: a retrospective observational study. EJNMMI Phys. 2021;8(1):31.\u003c/li\u003e\n \u003cli\u003eUsmani S, Marafi F, Ahmed N, Esmail A, Al Kandari F, Van den Wyngaert T. Diagnostic Challenge of Staging Metastatic Bone Disease in the Morbidly Obese Patients: A Primary Study Evaluating the Usefulness of 18F-Sodium Fluoride (NaF) PET-CT. Clin Nucl Med. 2017;42(11):829-36.\u003c/li\u003e\n \u003cli\u003eSanaat A, Shiri I, Arabi H, Mainta I, Nkoulou R, Zaidi H. Deep learning-assisted ultra-fast/low-dose whole-body PET/CT imaging. Eur J Nucl Med Mol Imaging. 2021;48(8):2405-15.\u003c/li\u003e\n \u003cli\u003eXu L, Yang R, Li RS, Liu RC, Meng QL, Wang F. Investigate the quantification accuracy of small lesions in oncological\u0026nbsp;18F-FDG PET/CT using a deep progressive learning reconstruction method. BMC Med Imaging. 2026;26(1):84.\u003c/li\u003e\n \u003cli\u003eLv Y, Xi C. PET image reconstruction with deep progressive learning. Phys Med Biol. 2021;66(10).\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eWang T, Qiao W, Wang Y, Wang J, Lv Y, Dong Y, et al. Deep progressive learning achieves whole-body low-dose\u0026nbsp;18F-FDG PET imaging. EJNMMI Phys. 2022;9(1):82.\u003c/li\u003e\n \u003cli\u003eYang H, Chen S, Qi M, Chen W, Kong Q, Zhang J, et al. Investigation of PET image quality with acquisition time/bed and enhancement of lesion quantification accuracy through deep progressive learning. EJNMMI Phys. 2024;11(1):7.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eObesity: preventing and managing the global epidemic. Report of a WHO consultation. World Health Organ Tech Rep Ser. 2000;894:i-xii, 1-253.\u003c/li\u003e\n \u003cli\u003eSonni I, Baratto L, Park S, Hatami N, Srinivas S, Davidzon G, et al. Initial experience with a SiPM-based PET/CT scanner: influence of acquisition time on image quality. EJNMMI Phys. 2018;5(1):9.\u003c/li\u003e\n \u003cli\u003eAdams MC, Turkington TG, Wilson JM, Wong TZ. A systematic review of the factors affecting accuracy of SUV measurements. AJR Am J Roentgenol. 2010;195(2):310-20.\u003c/li\u003e\n \u003cli\u003eMesserli M, Stolzmann P, Egger-Sigg M, Trinckauf J, D\u0026apos;Aguanno S, Burger IA, et al. Impact of a Bayesian penalized likelihood reconstruction algorithm on image quality in novel digital PET/CT: clinical implications for the assessment of lung tumors. EJNMMI Phys. 2018;5(1):27.\u003c/li\u003e\n \u003cli\u003eMcDermott GM, Chowdhury FU, Scarsbrook AF. Evaluation of noise equivalent count parameters as indicators of adult whole-body FDG-PET image quality. Ann Nucl Med. 2013;27(9):855-61.\u003c/li\u003e\n \u003cli\u003eAhn S, Ross SG, Asma E, Miao J, Jin X, Cheng L, et al. Quantitative comparison of OSEM and penalized likelihood image reconstruction using relative difference penalties for clinical PET. Phys Med Biol. 2015;60(15):5733-51.\u003c/li\u003e\n \u003cli\u003eZan K, Duan Y, Zhao M, Li H, Cui X, Chai L, et al. Performance of the Iterative OSEM and HYPER Algorithm for Total-body PET at SUVmax with a Low 18F-FDG Activity, a Short Acquisition Time and Small Lesions. Curr Med Imaging. 2024;20:e15734056274225..\u003c/li\u003e\n \u003cli\u003eHirji H, Sullivan K, Lasker I, Sharif MS, Nunes A, Shepherd C, et al. Effect of PET Image Reconstruction Techniques on Unexpected Aorta Uptake. Mol Imaging Radionucl Ther. 2019;28(1):1-7.\u003c/li\u003e\n \u003cli\u003eIwano S, Ito S, Tsuchiya K, Kato K, Naganawa S. What causes false-negative PET findings for solid-type lung cancer? Lung Cancer. 2013;79(2):132-6.\u003c/li\u003e\n \u003cli\u003eAdler S, Seidel J, Choyke P, Knopp MV, Binzel K, Zhang J, et al. Minimum lesion detectability as a measure of PET system performance. EJNMMI Phys. 2017;4(1):13.\u003c/li\u003e\n \u003cli\u003eKhalaf M, Abdel-Nabi H, Baker J, Shao Y, Lamonica D, Gona J. Relation between nodule size and 18F-FDG-PET SUV for malignant and benign pulmonary nodules. J Hematol Oncol. 2008;1:13.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"PET/CT, Deep progressive reconstruction, Image quality, Body mass index, Breast cancer","lastPublishedDoi":"10.21203/rs.3.rs-9025681/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9025681/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eTo evaluate the efficacy of Deep Progressive Reconstruction (DPR) versus Ordered Subset Expectation Maximization (OSEM) algorithms in enhancing ^18F-FDG PET image quality across different body mass index (BMI) strata in patients with breast cancer.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis retrospective study included patients with breast cancer who underwent diagnostic ^18F-FDG PET/CT at the China-Japan Union Hospital of Jilin University between June and December 2023. Whole-body PET/CT was performed using the United Imaging uMI 780 system, with images reconstructed using three algorithms: DPR, OSEM2 (two iterations), and OSEM3 (three iterations). Patients were stratified into four BMI subgroups: underweight, normal weight, overweight, and obese. Two board-certified nuclear radiologists independently and blindly assessed image quality, noise level, and lesion conspicuity using a 5-point Likert scale. Quantitative metrics included maximal lesion diameter, maximum standardized uptake value (SUVmax), peak SUV (SUVpeak), liver signal-to-noise ratio (LSNR), tumor-to-background ratio (T/N), and liver SUV standard deviation (SUVSD). Statistical analyses were performed with one-way analysis of variance (ANOVA) and Mann\u0026ndash;Whitney U tests to compare the three reconstruction methods across BMI subgroups and lesion sizes.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eA total of 118 female patients with 188 measurable lesions were analyzed. DPR-reconstructed PET images achieved significantly higher quality scores and lower noise levels than OSEM2 and OSEM3 (all P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Rising BMI significantly worsened image quality and noise scores in OSEM2/OSEM3 (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05), with OSEM3 in obese patients failing to meet diagnostic criteria (quality score: 2.6; noise score: 2.7). DPR reconstruction yielded significantly higher SUVmax, SUVpeak, LSNR, and T/N values, while producing markedly lower SUVSD compared to OSEM methods (all P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Compared with OSEM3, DPR conferred greater improvements in SUVmax, SUVpeak, and T/N in low-weight patients (P\u0026thinsp;=\u0026thinsp;0.011, 0.048, and 0.014, respectively), and larger gains in SUVSD and LSNR in high-weight patients (P\u0026thinsp;=\u0026thinsp;0.003 and \u0026lt;\u0026thinsp;0.001, respectively). Inverse correlations were observed between lesion size and DPR improvements versus OSEM2 for SUVmax (r=\u0026ndash;0.298), SUVpeak (r=\u0026ndash;0.243), and T/N (r=\u0026ndash;0.326) (all P\u0026thinsp;\u0026lt;\u0026thinsp;0.01), and versus OSEM3 for SUVmax (r=\u0026ndash;0.213, P\u0026thinsp;\u0026lt;\u0026thinsp;0.01). When compared with OSEM2, DPR provided greater benefits for sub-2 cm lesions (+\u0026thinsp;12% SUVmax, +\u0026thinsp;6% SUVpeak, +\u0026thinsp;17% T/N) than for larger lesions (+\u0026thinsp;5%, +\u0026thinsp;1%, +\u0026thinsp;6%; all P\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eDPR significantly improved ^18F-FDG PET image quality, noise suppression, and lesion conspicuity compared with OSEM. These advantages were most pronounced in patients with higher BMI and in small lesions, highlighting the potential of DPR to enhance diagnostic performance in challenging patient subgroups.\u003c/p\u003e","manuscriptTitle":"Efficacy Study of Deep Progressive Reconstruction Algorithm in Enhancing 18F-FDG PET Image Quality for Breast Cancer Patients with Different Body Mass Index","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-09 16:38:51","doi":"10.21203/rs.3.rs-9025681/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-01T08:21:56+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"88992278663209737787933666012259464066","date":"2026-04-09T13:52:58+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-03T10:40:36+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-10T09:00:15+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-09T14:21:08+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-09T14:20:13+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2026-03-04T04:21:00+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"6eecc634-3a40-4f84-a5cf-554f236c9b92","owner":[],"postedDate":"April 9th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-01T08:21:56+00:00","index":58,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-09T16:38:56+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-09 16:38:51","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9025681","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9025681","identity":"rs-9025681","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}