A total of 23 patients (female/male: 8/15, age: 25–75 years) were finally included in the study with two excluded participants. The demographic and clinical information of the enrolled patients are listed in Table 1. Among them, 4 patients were scanned for initial diagnosis or staging, 6 patients were for treatment monitoring, and the remained 13 patients were for recurrence detection.

In the semi-quantitative analysis, the uptake and noise in the liver were first analyzed. The SUVmean of liver in FB, BH_15 and BH_20 PET demonstrated good agreement with RG PET, indicating a high level of accuracy in PET quantification, as illustrated in Fig. 1. In comparison to FB PET, the image quality of RG, BH_15 and BH_20 PET was significantly degraded with a reduced COV value (p < 0.05, p < 0.001, and p < 0.001, respectively). This image quality degradation was primarily attributed to the limited counts acquired from the short time. Subsequently, the clinical acceptability of these PET images was assessed through visual analysis.

Bland-Altman plots of liver SUVmean in FB, BH_15, and BH_20 PET showed good agreement with that in RG PET

A total of 78 lesions were included in the analysis, and two out of 78 lesions were unsuccessfully segmented and recorded only with SUVmax information. The lesion SUVmax in RG, BH_15 and BH_20 PET were all found significantly higher than that in FB PET (p < 0.01, p < 0.001, and p < 0.001, respectively), as illustrated in Fig. 2A. Furthermore, both the BH groups showed significantly increased lesion SUVmax compared to RG PET, suggesting superior capability in elevating the lesion SUV measurements (p < 0.001). Nevertheless, no significant difference was observed between both BH PET images (p = 1.0). In addition, %SUV-BH15 and %SUV-BH20 proved significantly higher than %SUV-RG, indicating a more robust capability in recovering SUV measurements (both p < 0.001, Fig. 2B). Notably, there was no significant difference between %SUV-BH15 and %SUV-BH20 (p = 1.0). Figure 3 illustrates PET images of a 25-year-old female patient diagnosed with pancreatic NET (G1) with bone and liver metastases. As observed in the MIP images (upper row), lesions in the upper abdomen in FB PET are blurred with severe distortion, while these distortion can be well recovered in RG and both BH PET images. In the bottom row, RG PET and both BH PET can better image the lesions compared to FB PET as indicated with a red hollow arrow. In addition, an increase in the lesion SUVmax was found as 13.04 for FB PET, 19.75 for RG PET, 25.85 for BH_15 PET, and 25.09 for BH_20 PET.

Boxplots of (A) lesion SUVmax and (B) percentage difference of lesion SUVmax (%SUV) compared to the FB PET. Compared to FB PET, lesion SUVmax in RG, BH_15 and BH_20 groups was significantly higher. The percentage difference of lesion SUVmax in BH groups also showed significantly difference compared to RG group. However, there was no significant difference in lesion SUVmax and percentage difference of lesion SUVmax between BH groups. SUVmax, the maximum of standardized uptake value; %SUV, percentage difference in lesion SUVmax compared to FB PET; * p < 0.05; ** p < 0.01; *** p < 0.001; ns, no significant difference

PET images of a 25-year-old female patients diagnosed with pancreatic NET (G1) with liver and bone metastases. Severe distortion of lesions can be observed in both MIP and MPR images in FB PET, while these distortion can be well recovered in RG and both BH PET. NET, neuroendocrine tumor; MIP, maximum intensity projection; MPR, multi planar reconstruction; RG, respiratory gating; BH, breath hold

To investigate image contrast and lesion conspicuity, we employed TBR as a metric and observed a significant elevation in TBR values for RG, BH_15 and BH_20 PET compared to FB PET (all p < 0.001, Fig. 4). There was a significant enhanced image contrast in both BH PET images compared with RG PET images (both p < 0.001), and no significance was found between both BH PET images (p = 1.0).

Comparison of TBR between FB, RG, BH_15 and BH_20 PET. Compared with FB PET, the TBR in RG, BH_15 and BH_20 PET were significantly higher, indicating a higher image contrast and a better lesion conspicuity. BH PET groups further showed a significantly higher TBR than that in RG PET, but no significance was found between both BH PET groups. TBR, target-to-background ratio; *** p < 0.001; ns, no significant difference

With regard to MTV, the mean of RG PET exhibited a comparable value with that in FB PET without significant difference (p = 0.094), but higher than those in both BH PET (both p < 0.001), as illustrated in Fig. 5A. Likewise, the MTV between both BH PET images was not significantly different (p = 0.750, Fig. 5A). Regarding the percentage difference in MTV, we found that both BH PET had a significantly lower value than that in RG PET (both p < 0.001, Fig. 5B), indicating their capacity of BH PET to further reduce the MTV compared to RG PET. Notably, there was no significant difference in the percentage difference in MTV between both BH PET (p = 1.0).

Boxplots of (A) lesion MTV and (B) percentage difference in MTV compared to FB PET. RG, BH_15 and BH_20 PET showed a significantly lower MTV than that in FB PET. Both BH PET had significantly reduced MTV compared to RG PET, and no significance was found between BH groups. MTV, metabolic tumor volume; %MTV, percentage difference in MTV compared to FB PET; **p < 0.01; *** p < 0.001; ns, no significant difference; RG, respiratory gating; BH, breath hold

In the subgroup analysis categorized by the tumor location, we observed varying statistical significances in liver tumors (n = 60), pancreatic tumors (n = 6) and lymph nodes (n = 8). Regarding the lesion SUVmax, both BH PET showed significantly increase compared to RG and FB PET in liver and pancreatic tumors (all p < 0.05), but no significant difference in lymph nodes (p = 1.0 for BH_15 and BH_20, Fig. 6). In addition, compared to FB PET, RG PET only significantly enhanced the SUVmax measurement in liver tumors (p < 0.01), but no statistical difference was found in pancreatic tumors and lymph nodes (both p = 1.0). In the analysis of MTV, there were significant differences between both BH PET and RG PET images in liver tumors (both p < 0.001), but no significant differences in pancreatic tumors (p = 0.650 and 0.630) and lymph nodes (both p = 1.0). However, both BH PET significantly reduced MTV compared to FB PET in liver tumors (both p < 0.001), but still showed comparable measurements in pancreatic tumors and lymph nodes (both p = 1.0). With regard to TBR, RG PET and both BH PET showed significantly improved measurements compared to FB PET in liver (all p < 0.001) and pancreatic tumors (all p < 0.05), but not in lymph nodes. Notably, no significant difference was observed between both BH PET images in the above analysis.

Subgroup analysis of (A) lesion SUVmax, (B) MTV, and (C) target-to-background ratio (left: liver; middle: pancreas; right: lymph node) for different tumor location. Differences in statistical significance can be found in different tumor locations. SUVmax, the maximum of standardized uptake value; MTV, metabolic tumor volume; * p < 0.05; ** p < 0.01; *** p < 0.001; ns, no significant difference; RG, respiratory gating; BH, breath hold

The subsequent subgroup analysis was performed based on the tumor size, with all included lesions categorized into two groups with a cutoff value of 1.0 cm3: a large lesion group (L-group) and a small lesion group (S-group) measured in RG PET images. In the L-group (n = 18), no significant difference was observed in lesion SUVmax, MTV and TBR between FB and RG PET (all p = 1.0), suggesting a limited effectiveness of the RG technique in larger lesions (Fig. 7). However, SUVmax in BH_20 PET was significantly higher (p < 0.01) while lesion SUVmax in BH_15 PET was slightly elevated; MTV was significantly reduced (both p < 0.01); and the TBR was significantly improved (p < 0.01, and < 0.001, respectively), compared to FB PET. This implies RG PET yielded comparable quantitative measurements to FB PET in lesions with larger volumes, while both BH PET showed significant or slight enhancement in these measurements. Conversely, in the S-group (n = 58), MTV in RG and both BH PET were significantly lower than that in FB PET (p < 0.05, 0.01 and 0.001, respectively), while SUVmax were significantly elevated (p < 0.01, 0.001 and 0.001, respectively). Consequently, TBR in RG and both BH PET showed a significant improvement compared to FB PET (all p < 0.001). Moreover, in these two subgroups, no significant difference was found in SUVmax, MTV and TBR between both BH PET groups, indicating comparable quantitative measurements regardless of tumor size (L-group: p = 0.462, 0.395, and 0.356; S-group: p = 0.483, 0.265 and 0.736, respectively).

Subgroup analysis of (A) lesion SUVmax, (B) MTV, and (C) target-to-background ratio (left: lesions with large volume; right: lesions for small volume) for different lesion volume. Compared to FB PET, both BH PET can significantly increase lesion SUVmax, TBR while reduce MTV in both groups. There was no significant difference in the three metrics between both BH PET images. SUVmax, the maximum of standardized uptake value; MTV, metabolic tumor volume; * p < 0.05; ** p < 0.01; *** p < 0.001; ns, no significant difference; RG, respiratory gating; BH, breath hold

Lesion uptake served as another criteria for categorizing lesions into two subgroups: a high uptake group (HU-group) and a low uptake group (LU-group), using a cutoff value of 16.3 measured in RG PET images. In the HU-group (n = 32), both BH PET exhibited a significantly higher SUVmax and lower MTV compared with FB PET (all p < 0.05), as illustrated in Fig. 8. However, RG PET showed comparable SUV and MTV with FB PET (both p = 1.0). The TBR in RG PET was significantly higher than in FB PET (p < 0.05), and lower than in both BH PET (both p < 0.001). No statistical significance was found in SUVmax, MTV and TBR between both BH groups. In contrast, in the LU-group low uptake group (n = 46), there was a significant difference in SUVmax and TBR between RG and FB PET (p < 0.01, and 0.001, respectively), but no difference in MTV. However, both BH groups exhibited a significantly higher SUVmax (both p < 0.001), lower MTV (both p < 0.01), and higher TBR (both p < 0.001) compared to FB PET. Similarly, no statistical significance was found in these three metrics between both BH groups (all p = 1.0).

Subgroup analysis of (A) lesion SUVmax, (B) MTV, and (C) target-to-background ratio (left: lesions with higher uptake; right: lesions for lower uptake) for different lesion SUVmax. BH PET can significantly increase lesion SUVmax, TBR while reduce MTV regardless of lesion SUVmax. There was no significant difference in the three metrics between both BH PET images. SUVmax, the maximum of standardized uptake value; MTV, metabolic tumor volume; * p < 0.05; ** p < 0.01; *** p < 0.001; ns, no significant difference; RG, respiratory gating; BH, breath hold

The overall inter-reader agreement in the visual analysis demonstrated a Cohen’s kappa of 0.719, representing a substantial agreement. The mean scores ± SD for FB, RG, BH_15, and BH_20 PET were as follows: for respiratory motion artifacts, 2.61 ± 1.16, 1.61 ± 0.72, 1.17 ± 0.39, and 1.13 ± 0.34; for overall diagnostic confidence, 1.56 ± 0.73, 1.09 ± 0.29, 1.13 ± 0.34, and 1.04 ± 0.21; and for PET and CT alignment, 1.87 ± 0.81, 1.87 ± 0.76, 1.52 ± 0.66, and 1.65 ± 0.78. Regarding the respiratory motion artifacts, RG and both BH PET exhibited significantly fewer artifacts compared to FB PET (p < 0.01, 0.001, and 0.001, respectively), as illustrated in Fig. 9. Both BH PET showed significantly reduced respiratory motion artifacts compared to RG PET (both p < 0.05), suggesting a superior performance of the BH approach in mitigating respiratory motion. No significant difference was observed between RG and both BH PET (both p = 1.0) in terms of diagnostic confidence, indicating a comparable performance for clinical use. However, all of them showed significantly higher diagnostic confidence compared to FB PET (all p < 0.05). As shown in Fig. 10, a high uptake lesion (arrow) was found in a patient with rectum NET for treatment monitoring and suspected of a metastatic lymph node or a metastasis of gastric antrum in routine FB PET/CT images. However, the diagnostic confidence can be improved in both BH PET/CT images. With regards to PET and CT alignment, no significant difference was observed between PET series. Likewise, there was no significant difference in the three perspectives between both BH PET images (p = 1.0, 0.785, and 1.0, respectively).

Scores of visual analysis in the three perspectives: (A) respiratory motion artifacts; (B) diagnostic confidence; (C) PET and CT alignment. Regarding the respiratory motion artifacts and diagnostic confidence, BH PET exhibited significantly better performance than FB PET. However, there was no significant difference between FB, RG and both BH PET regarding the PET and CT alignment. ** p < 0.01; *** p < 0.001; ns, no significant difference; RG, respiratory gating; BH, breath hold

PET/CT images of a 57-year-old male patient with rectum NET. The diagnostic confidence on the metastasis (arrow) was improved in BH PET/CT compared with FB PET/CT. NET, neuroendocrine tumor; MIP, maximum intensity projection; RG, respiratory gating; BH, breath hold

We employed RG PET as a reference for evaluating lesion detectability. In a patient-base analysis, FB PET detected additional lesions (3/23) and both BH PET detected even additional lesions (10/23). Missed lesions were observed in both BH PET (3/23), while FB PET missed lesions in additional three patients (6/23 in total). Furthermore, we found that these additional findings and omissions didn’t alter the clinical management in the present study due to the nature of multiple metastases in NETs.