Objective:To explore the feasibility of high-level deep learning image reconstruction (DLIR-H) algorithm in dynamic myocardial perfusion (CTP) and coronary CT angiography (CCTA) extraction.Methods:From October 2021 to October 2022, 41 patients with confirmed or suspected coronary heart disease who underwent traditional CCTA and dynamic CTP examinations on GE Apex CT were prospectively collected. Traditional CCTA used 100 kVp tube voltage scan and DLIR-H to reconstruct the original image, while dynamic CTP used 80 kVp tube voltage scan to reconstruct the original image with ASiR-V 100% and DLIR-H, respectively. Comparing subjective and objective scores and myocardial blood flow (MBF) values?of rest and stress CTP between ASiR-V100% and DLIR-H. Subjective and objective scores, as well as stenosis degree and coronary CT blood flow reserve fraction (CT-FFR) value were analyzed on a vessel basis, and the image quality and diagnostic performance of traditional CCTA and single-phase CCTA (SP-CCTA) extracted under DLIR-H CTP were compared. Statistical analysis were performed using paired t test, Wilcoxon signed-rank test and χ2 test. Results:In the subjective image quality analysis of resting and stress CTP, DLIR-H was improved compared with ASiR-V100%, and the difference was statistically significant (all P<0.05). There was no significant difference in MBF values obtained by the two reconstruction methods in the assessment of quantitative myocardial perfusion ( P>0.05). Compared with traditional CCTA, the vascular CT value of SP-CCTA increased by 15.12%, the noise value increased by 32.27%, and the subjective score was also slightly lower (4.23±0.05). However, there were no statistically significant differences in total plaque volume, maximum stenosis degree, and number of CT-FFR positive vessels between SP-CCTA and traditional CCTA (all P>0.05). Conclusion:The deep learning reconstruction algorithm can not only improve the quality of the original image to a certain extent on dynamic CTP, but also extract high-quality single-phase CCTA to meet clinical diagnosis and realize a “one-stop” dynamic myocardial perfusion examination, which will help simplify the examination process, reduce contrast agent and radiation doses in the future.

Purpose To evaluate the effectiveness of deep learning image reconstruction(DLIR)algorithms in reconstructing lumbar vertebrae images from abdominal CT scans,aiming to reduce radiation dose and eliminate the need for repeat lumbar CT examinations.Materials and Methods A retrospective collection was conducted from March to May 2024 in the First Affiliated Hospital of Tianjin University of Traditional Chinese Medicine.Thirty-two patients who underwent both abdominal and lumbar CT scans in a supine head-first position were enrolled.The abdominal CT(DLIR group)utilized a tube voltage of 120 kVp and a current of 200 mA with high-intensity DLIR for lumbar reconstruction.The standard lumbar CT(lumbar group)used the same voltage with a tube current of 260 mA and was reconstructed using 60%weighted adaptive statistical iterative reconstruction.Objective assessments was used to measure the CT values,noise(standard deviation,SD value),signal-to-noise ratio and contrast-to-noise ratio(excluding adipose tissue)at the third lumbar vertebral pedicle level and the L2/L3 intervertebral disc level for muscle,adipose tissue,cancellous bone,intervertebral discs,dura mater and cortical bone.Subjective assessments employed a five-point scale to evaluate image contrast,noise and sharpness.Results The volume CT dose index in lumbar group and DLIR group were 15.25 mGy and 11.74 mGy,respectively.There was no statistical difference in CT values between the structures of both groups(all P>0.05).Compared with the lumbar group,the DLIR group showed significant reductions in SD values across the measured tissues by 31.09%,35.66%,13.48%,27.82%,24.93%and 15.09%(t=5.09-7.21,all P<0.05).The signal-to-noise ratio improved by 36.40%,52.31%,16.56%,34.13%,38.39%and 18.81%,and the contrast-to-noise ratio improved by 51.70%,51.32%,36.24%,34.47%and 53.56%(t=-9.58--4.23,all P<0.001).The DLIR group significantly outperformed the lumbar group in image contrast[4.45(4.00,5.00)points vs.4.75(4.00,5.00)points],image noise[4.06(4.00,4.00)points vs.4.39(4.00,5.00)points],and spatial resolution of fine structures[4.00(4.00,4.00)points vs.4.27(4.00,5.00)points](Z=-3.80,-4.38,-3.55,all P<0.001).Conclusion Using high-intensity DLIR for abdominal examinations can achieve high-quality lumbar CT images with a 25%reduction in radiation dose,enabling simultaneous abdominal and lumbar scanning in a single session.

Objective To observe the value of deep learning image reconstruction(DLIR)algorithm in brain CT perfusion(CTP)using a protocol of 70 kVp and 40 ml contrast agent dose.Methods Totally 105 patients with suspected acute ischemic stroke(AIS)were prospectively enrolled and randomly divided into 3 groups,who underwent standard dose CTP scanning with 80 kVp and 150 mA combined with reconstruction as adaptive statistic iterative reconstruction V(ASIR-V)at 50％level(CN group,n=35),low dose(LD)scanning with 70 kVp and 100 mA combined with DLIR reconstruction at the highest level(DLIR-H)(LD group,n=35),or ultra-low dose(ULD)scanning with 70 kVp and 70 mA combined with DLIR-H reconstruction(ULD group,n=35).Radiation doses were compared among 3 groups.CT values and standard deviations(SDCT)of ROI of gray matter and white matter in the frontal,parietal and temporal lobes were measured.Signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)between gray and white matter were calculated and compared among groups.Then pseudo-color images of cerebral blood volume(CBV),cerebral blood flow(CBF),mean transit time(MTT)and time to maximum of the tissue residual function(Tmax)were generated.The imaging quality of CTP pseudo-color images was evaluated,and the compatibility of the subjective scores within every kind of CTP pseudo-color images were assessed using Kappa test.Quantitative perfusion parameters were measured and compared among groups.Results Compared with CN group,both LD and ULD groups demonstrated significantly reduced volume CT dose index(CTDIvol),dose-length product(DLP)and effective dose(ED)(all adjusted P＜0.05).In ULD group,SDCT of white matter in frontal,parietal and temporal lobes were higher than those in CN group,and SDCT of white matter in parietal lobe was also higher than that in LD group(all adjusted P＜0.05).No significant difference of SDCT of gray matter was observed among groups(all P＞0.05).SNR of white matter in parietal and temporal lobes in both LD and ULD groups were lower than those in CN group(all P＜0.05),while no significant difference of SNR of white matter in frontal lobe,nor of gray matter in frontal,parietal and temporal lobes was found among groups(all P＞0.05).CNR of gray and white matter in the frontal,parietal and temporal lobes were not significantly different among groups(all P＞0.05).High consistency of inter-observer subjective scores of CBV maps,CBF maps and Tmax maps(Kappa of 0.623,0.644 and 0.638,respectively)were noticed,which of MTT maps had moderate consistency(Kappa=0.560).No significant difference of intra-obsever subjective scores of CTP pseudo-color images was found among groups(all P＞0.05).CBV,CBF,MTT and Tmax values of gray and white matter in frontal,parietal and temporal lobes were not significantly different among groups(all P＞0.05).Conclusion DLIR algorithm applicated in low radiation dose and reduced contrast agent dosage might ensure imaging quality.

Purpose To evaluate the effectiveness of deep learning image reconstruction(DLIR)algorithms in reconstructing lumbar vertebrae images from abdominal CT scans,aiming to reduce radiation dose and eliminate the need for repeat lumbar CT examinations.Materials and Methods A retrospective collection was conducted from March to May 2024 in the First Affiliated Hospital of Tianjin University of Traditional Chinese Medicine.Thirty-two patients who underwent both abdominal and lumbar CT scans in a supine head-first position were enrolled.The abdominal CT(DLIR group)utilized a tube voltage of 120 kVp and a current of 200 mA with high-intensity DLIR for lumbar reconstruction.The standard lumbar CT(lumbar group)used the same voltage with a tube current of 260 mA and was reconstructed using 60%weighted adaptive statistical iterative reconstruction.Objective assessments was used to measure the CT values,noise(standard deviation,SD value),signal-to-noise ratio and contrast-to-noise ratio(excluding adipose tissue)at the third lumbar vertebral pedicle level and the L2/L3 intervertebral disc level for muscle,adipose tissue,cancellous bone,intervertebral discs,dura mater and cortical bone.Subjective assessments employed a five-point scale to evaluate image contrast,noise and sharpness.Results The volume CT dose index in lumbar group and DLIR group were 15.25 mGy and 11.74 mGy,respectively.There was no statistical difference in CT values between the structures of both groups(all P>0.05).Compared with the lumbar group,the DLIR group showed significant reductions in SD values across the measured tissues by 31.09%,35.66%,13.48%,27.82%,24.93%and 15.09%(t=5.09-7.21,all P<0.05).The signal-to-noise ratio improved by 36.40%,52.31%,16.56%,34.13%,38.39%and 18.81%,and the contrast-to-noise ratio improved by 51.70%,51.32%,36.24%,34.47%and 53.56%(t=-9.58--4.23,all P<0.001).The DLIR group significantly outperformed the lumbar group in image contrast[4.45(4.00,5.00)points vs.4.75(4.00,5.00)points],image noise[4.06(4.00,4.00)points vs.4.39(4.00,5.00)points],and spatial resolution of fine structures[4.00(4.00,4.00)points vs.4.27(4.00,5.00)points](Z=-3.80,-4.38,-3.55,all P<0.001).Conclusion Using high-intensity DLIR for abdominal examinations can achieve high-quality lumbar CT images with a 25%reduction in radiation dose,enabling simultaneous abdominal and lumbar scanning in a single session.

Objective To observe the value of deep learning image reconstruction(DLIR)algorithm in brain CT perfusion(CTP)using a protocol of 70 kVp and 40 ml contrast agent dose.Methods Totally 105 patients with suspected acute ischemic stroke(AIS)were prospectively enrolled and randomly divided into 3 groups,who underwent standard dose CTP scanning with 80 kVp and 150 mA combined with reconstruction as adaptive statistic iterative reconstruction V(ASIR-V)at 50％level(CN group,n=35),low dose(LD)scanning with 70 kVp and 100 mA combined with DLIR reconstruction at the highest level(DLIR-H)(LD group,n=35),or ultra-low dose(ULD)scanning with 70 kVp and 70 mA combined with DLIR-H reconstruction(ULD group,n=35).Radiation doses were compared among 3 groups.CT values and standard deviations(SDCT)of ROI of gray matter and white matter in the frontal,parietal and temporal lobes were measured.Signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)between gray and white matter were calculated and compared among groups.Then pseudo-color images of cerebral blood volume(CBV),cerebral blood flow(CBF),mean transit time(MTT)and time to maximum of the tissue residual function(Tmax)were generated.The imaging quality of CTP pseudo-color images was evaluated,and the compatibility of the subjective scores within every kind of CTP pseudo-color images were assessed using Kappa test.Quantitative perfusion parameters were measured and compared among groups.Results Compared with CN group,both LD and ULD groups demonstrated significantly reduced volume CT dose index(CTDIvol),dose-length product(DLP)and effective dose(ED)(all adjusted P＜0.05).In ULD group,SDCT of white matter in frontal,parietal and temporal lobes were higher than those in CN group,and SDCT of white matter in parietal lobe was also higher than that in LD group(all adjusted P＜0.05).No significant difference of SDCT of gray matter was observed among groups(all P＞0.05).SNR of white matter in parietal and temporal lobes in both LD and ULD groups were lower than those in CN group(all P＜0.05),while no significant difference of SNR of white matter in frontal lobe,nor of gray matter in frontal,parietal and temporal lobes was found among groups(all P＞0.05).CNR of gray and white matter in the frontal,parietal and temporal lobes were not significantly different among groups(all P＞0.05).High consistency of inter-observer subjective scores of CBV maps,CBF maps and Tmax maps(Kappa of 0.623,0.644 and 0.638,respectively)were noticed,which of MTT maps had moderate consistency(Kappa=0.560).No significant difference of intra-obsever subjective scores of CTP pseudo-color images was found among groups(all P＞0.05).CBV,CBF,MTT and Tmax values of gray and white matter in frontal,parietal and temporal lobes were not significantly different among groups(all P＞0.05).Conclusion DLIR algorithm applicated in low radiation dose and reduced contrast agent dosage might ensure imaging quality.

Objective:To explore the feasibility of high-level deep learning image reconstruction (DLIR-H) algorithm in dynamic myocardial perfusion (CTP) and coronary CT angiography (CCTA) extraction.Methods:From October 2021 to October 2022, 41 patients with confirmed or suspected coronary heart disease who underwent traditional CCTA and dynamic CTP examinations on GE Apex CT were prospectively collected. Traditional CCTA used 100 kVp tube voltage scan and DLIR-H to reconstruct the original image, while dynamic CTP used 80 kVp tube voltage scan to reconstruct the original image with ASiR-V 100% and DLIR-H, respectively. Comparing subjective and objective scores and myocardial blood flow (MBF) values?of rest and stress CTP between ASiR-V100% and DLIR-H. Subjective and objective scores, as well as stenosis degree and coronary CT blood flow reserve fraction (CT-FFR) value were analyzed on a vessel basis, and the image quality and diagnostic performance of traditional CCTA and single-phase CCTA (SP-CCTA) extracted under DLIR-H CTP were compared. Statistical analysis were performed using paired t test, Wilcoxon signed-rank test and χ2 test. Results:In the subjective image quality analysis of resting and stress CTP, DLIR-H was improved compared with ASiR-V100%, and the difference was statistically significant (all P<0.05). There was no significant difference in MBF values obtained by the two reconstruction methods in the assessment of quantitative myocardial perfusion ( P>0.05). Compared with traditional CCTA, the vascular CT value of SP-CCTA increased by 15.12%, the noise value increased by 32.27%, and the subjective score was also slightly lower (4.23±0.05). However, there were no statistically significant differences in total plaque volume, maximum stenosis degree, and number of CT-FFR positive vessels between SP-CCTA and traditional CCTA (all P>0.05). Conclusion:The deep learning reconstruction algorithm can not only improve the quality of the original image to a certain extent on dynamic CTP, but also extract high-quality single-phase CCTA to meet clinical diagnosis and realize a “one-stop” dynamic myocardial perfusion examination, which will help simplify the examination process, reduce contrast agent and radiation doses in the future.

Objective:To understand the etiological characteristics of 2 serotypes of Salmonella strains isolated from a foodborne disease outbreak. Methods:A total of 11 anal swabs of the cases, 13 suspected contaminated food and 10 environmental samples were collected from a foodborne disease outbreak occurred on September 8, 2022 in a school. The anal swabs were enriched with selenite brilliant green enrichment broth (SBG) and brain heart infusion broth (BHI) respectively. PCR detection and culture of common intestinal pathogens were carried out. The suspected food samples were tested according to national standards for food safety. Multiple suspected Salmonella colonies were obtained and selected for serotype determination and whole genome sequencing. Serotypes were determined based on the whole-genome sequence, and clustering analysis was performed based on core genome single nucleotide polymorphism (SNP). Results:The positive rates of Salmonella in anal swabs and suspected food samples were 9/11 and 5/13 respectively. Both Salmonella Uganda and Salmonella Idikan were isolated from 4 anal swabs and 4 suspected food samples. For the remaining samples, only Salmonella Uganda or Salmonella Idikan was isolated in each sample. The positive rate of Salmonella in 11 anal swabs of the cases after BHI enrichment for 12 h and 24 h were all 9/11 in real-time PCR, same to the culture results. Salmonella Uganda and Salmonella Idikan formed two independent and genetically distant lineages in the clustering tree based on core genome SNP, and 0-14 and 0-23 SNP were observed in Salmonella Uganda and Salmonella Idikan respectively. Conclusions:This foodborne disease outbreak was probably caused by Salmonella Uganda and Salmonella Idikan, which both exhibited strong genetic diversity. The PCR based pathogen screening strategy plus pathogen enrichment for cases' annal swabs can be used in the routine outbreak investigation.

Objective:To investigate the efficiency of deep learning image reconstruction (DLIR) algorithm in the image quality and detection of hypovascular hepatic metastases under low radiation doses in comparison with adaptive statistical iterative construction-V (ASiR-V).Methods:Fifty-six patients with suspected hypovascular hepatic metastases who needed abdominal enhanced CT scans were collected prospectively in the First Affiliated Hospital of Zhengzhou University from January to April 2021. The patients received conventional radiation dose with tube current-time products of 400 mA CT scans in the first venous phase, low-dose CT scans in the second venous phase, which were set as tube current-time products of 280 mA for group A (19 cases), 200 mA for group B (19 cases) and 120 mA for group C (18 case), respectively. The images of first venous phase and 3 groups of second venous phase were both reconstructed with ASiR-V60% and high-DLIR (DLIR-H). Quantitative parameters [image noise, liver and portal vein signal to noise ratio (SNR), contrast to noise ratio (CNR)] and qualitative parameters (overall image quality, lesion conspicuity, diagnostic confidence) were compared between ASiR-V60% and DLIR-H images, and the effective radiation dose (ED) and the lesion detectability of each group was recorded. The paired t test was used to compare quantitative parameters, whereas the Wilcoxon signed-rank test of paired data was used to compare qualitative parameters. Results:In the second venous phase, ED was (5.56±0.35) mSv in group A, (3.88±0.23) mSv in group B, and (2.42±0.23) mSv in group C, with a decrease of 30%, 50% and 70% compared with the first venous phase, respectively. Moreover, with the decrease of radiation dose, the noise gradually increased, and the CNR lesions, SNR liver and SNR portal vein all gradually decreased. DLIR-H images had statistically better quantitative scores than ASiR-V60% images when the same radiation dose was applied (all P<0.001). Furthermore, the qualitative parameters of each group decreased with the decrease of radiation dose. Under the same radiation dose, the overall image quality, lesion conspicuity and diagnostic confidence of DLIR-H were higher than those of ASiR-V60% (all P<0.001). All lesions [100% (84/84)] were detected by ASIR-V60% and DLIR-H in group A, 92.0% (75/81) in group B, and 88.0% (79/89) in group C. Conclusions:Compared with ASiR-V60%, DLIR-H could reduce image noise, improve overall image quality and lesion conspicuity of hypovascular hepatic metastases as well as increase diagnostic confidence under different radiation doses.