The present study involved two patient cohorts: an initial exploration cohort for the analysis of different acquisition times and a clinical validation cohort for the validation of the initial results and the optimization of acquisition times. The results for these two cohorts are presented below.
Initial exploration cohort
The demographic and clinical data of the patients are summarized in Table 1. We enrolled a total of 46 patients (34 men, 12 women) in the exploration cohort, with a mean age of 61.5 ± 8.8 years, a mean body mass index (BMI) of 24.1 ± 3.5 kg/m2, and a mean injected FDG dose of 1.88 ± 0.09 MBq/kg. The overall distribution of primary neoplasms in this cohort was as follows: lung neoplasms, 19 patients; colorectal cancer, 17 patients; liver cancer, 4 patients; biliary tract cancer, 2 patients; and stomach neoplasms, 4 patients. Lymph node metastases were suspected in 18 patients, while distant metastases were not present in any patient. The distribution of the primary tumors and lymph node lesions is shown in detail in Table 2.
Evaluation of image quality
Qualitative (subjective) image analyses were performed by two readers, and their findings showed excellent inter-rater agreement, with a weighted kappa of 0.848 (95% confidence interval [CI]: 0.800–0.897). The average qualitative scores for images with acquisition times of 1, 2, 3, 5, and 8 min (henceforth referred to as G1, G2, G3, G5, and G8, respectively) were 2.0 ± 0.2, 2.8 ± 0.3, 3.1 ± 0.2, 3.9 ± 0.3, and 4.1 ± 0.2, respectively (Table 3). Even for the G1 images, the average score indicated visually acceptable images with no need for rescanning. In the case of two patients, however, the G1 images were given a score of 1 point by both readers. The subjective scores of the G5 and G8 images were significantly higher than those of the other images (all p < 0.05). No significant difference in subjective scores was observed between the G5 and G8 images (p = 0.89). In addition, no distinct difference was identified between the G2 and G3 images (p > 0.99).
The results of the objective image-quality assessments are presented in Table 4. We calculated the average difference in the objective quality of the G1, G2, G3, G5, and G8 images from that of the G15 images (15-min acquisition time; Fig. 2) by subtracting the measurement results of G1, G2, G3, G5, and G8 by those of the G15 images (referred to as 1-15, 2-15, 3-15, 5-15, and 8-15 for the sake of simplicity). In general, images with longer acquisition times showed lower standardized uptake values (SUVs). The maximum SUV (SUVmax) of the liver was significantly lower in the G15 and G8 images than in the other images (all p < 0.05), but did not differ between the G15 and G8 images (p > 0.99). The SUVmax of the mediastinal blood pool was significantly lower in the G15, G8, and G5 images than in the G3, G2, and G1 (all p < 0.05), but did not differ between the G15 and G8 images (p = 0.13) or between the G8 and G5 images (p = 0.92). The average standard deviation (SD) of the regions of interest (ROIs) in the liver and mediastinal blood pool significantly differed between the G15 images and the other images (all p ≤ 0.001). The trend of the between-group differences in average SUVs was similar to the trend observed in the SUVs. Moreover, these differences were statistically significant among all groups (all p < 0.05), except for the differences in the blood pool SUVmax (p = 0.61) and SUVSD (p = 0.09) between the 8-15 and 5-15 groups.
Pathological examination confirmed a total of 75 lesions in the 46 patients in the exploration cohort. Of these, 7 lesions in 6 patients (2 liver lesions, 1 lung lesion, and 4 lymph node lesions) were not recognizable on G15 images. In all, 47 primary lesions and 21 suspicious lymph node metastases were detected on G15 images. With the G15 data as the reference, the lesion-detection rates were 85.3% (58/68) and 97.1% (66/68) for the G1 and G2 images, respectively, and 100% (68/68) for the remaining images. On the G1 images, 10 lesions from 6 patients were not identifiable, including 1 lesion in the liver and 9 lesions in the lymph nodes. The lesion-detection rate significantly differed between the G15 and G1 images (p = 0.001).
For the assessment of lesion conspicuity, a total of 56 out of 75 lesions were pathologically positive and included for analysis. The SUVmax, SUVpeak, and target-to-background ratio (TBR) of these 56 lesions were significantly lower on G15 images than on the remaining images (all p ≤ 0.04); moreover, these values did not significantly differ among the remaining images (all p ≥ 0.08; Table 5).
Clinical validation cohort
A total of 147 eligible patients (79 men, 68 women) with a mean age of 59.4 ± 12.1 years, mean body weight of 63.8 ± 11.7 kg, mean BMI of 23.6 ± 3.6 kg/m2, and mean injected FDG dose of 1.88 ± 0.10 MBq/kg were included in this dataset (Table 1). After integrating the pathological data, we included 240 lesions in the final analysis: 163 primary tumors, 69 suspicious lymph node metastases, and 8 distant metastases. Three patients had multiple primary tumors: one patient had small intestine cancer with schwannomas; one had bladder cancer with ureteral cancer, and another patient had liver cancer with gastrointestinal stromal tumor. Distant metastases were present in 4 patients, including 3 patients with colorectal cancer and liver metastasis, and 1 patient with breast cancer and bone metastasis. The distribution of all pathologically confirmed lesions in the validation cohort is shown in detail in Table 6.
Subjective image quality
The weighted kappa coefficient between the readers was 0.760 (95% CI: 0.723–0.797), which indicated a substantial degree of agreement. The subjective scores for the G2, G3, G5, G8, and Gs (acquisition time, 10 or 15 min) images were 2.9 ± 0.2, 3.0 ± 0.0, 3.5 ± 0.4, 4.0 ± 0.2, and 4.5 ± 0.4, respectively (Table 3). The scores for the G2 and G3 images were approximately 3 points or slightly lower than 3 points. Significant differences in these scores were observed between any two groups (all p < 0.05), except for G2 and G3 (p > 0.99).
Of the 240 lesions, 36 lesions were not clearly identified on Gs images: 11 liver lesions, 5 biliary tract lesions, 2 pancreatic lesions, 2 bladder lesions, 1 gallbladder lesion, and 15 lymph node lesions. Compared to the Gs images, the G2, G3, G5, and G8 images had lesion-detection rates of 89.7% (183/204), 94.1% (192/204), 99.0% (202/204), and 100% (204/204), respectively. The distribution of the 21 lesions that went undetected on G2 images was as follows: biliary tract (n = 4), liver (n = 2), pancreas (n = 1), stomach (n = 1), bladder (n = 1), small intestine (n = 1), lymph nodes (n = 10), and liver metastasis (n = 1). The distribution of lesions that were unidentifiable on G3 images was as follows: biliary tract (n = 2), bladder (n = 1), liver (n = 1), and lymph nodes (n = 8). The lesion-detection rates for the G5 and G8 images were not significantly lower than the rates for the Gs images (all p ≥ 0.50).