人工智能重建在头颅3D T1WI成像中的应用价值

The value of artificial intelligence reconstruction in cranial 3D T1 weighted imaging

目的探索人工智能(AI)重建方式在头颅3D T1WI图像(T1WI)中的应用。方法前瞻性收集2021年5月至2022年10月在南京鼓楼医院进行头颅磁共振检查的52例患者。每例患者均采集压缩感知因子为3和6的3D T1WI(CS3、CS6),同时进行人工智能重建,生成AI3及AI6图像。在四组图像中勾画感兴趣区,计算图像信噪比(SNR)及对比噪声比(CNR)。从灰白质对比、深部核团轮廓、舌肌肉噪声、视神经清晰度四个方面对不同序列图像进行定性评估。对图像的SNR及CNR差异采用单因素重复测量方差分析,采用Friedman检验比较定性分析结果差异。结果在相同CS因子下,AI重建图像SNR、CNR明显优于的常规CS重建图像[AI3_(SNR)(20.63±6.90)vs CS3_(SNR)(17.12±5.27),P<0.001;AI3_(CNR)(12.93±4.79)vs CS3_(CNR)(10.94±3.70),P<0.001;AI6_(SNR)(20.60±7.46)vs CS6_(SNR)(12.33±3.22),P<0.001;AI6CNR(12.84±5.20)vs CS6_(CNR)(7.66±2.43),P<0.001]。图像定性分析中,AI重建图像的评分明显高于常规重建图像[(AI3(11.56±0.72)vs CS3(11.29±0.75),P<0.001;AI6(11.67±0.61)vs CS6(7.51±0.84),P<0.001)]。AI6图像的SNR及CNR均高于CS3图像,但差异无统计学意义(P>0.05)。CS3、AI6图像的总体评分明显高于CS6图像(P<0.01)。结论与CS3相比,AI6在保证头颅3D T1WI图像分辨率的基础上,同时可以缩短将近一半的扫描时间。

Objective On the basis of the Compressed SENSE collection technology,the application of artificial intelligence reconstruction(AI)in the cranial 3D T1WI image(3D T1WI)was explored.Methods Fifty-two patients who underwent cranial magnetic resonance examination in Nanjing Drum Tower Hospital from May 2021 to October 2022 was prospectively collected.The se‐quence of 3D T1WI in each patient was collected with compressed sensing factors of 3 and 6(CS3,CS6),and artificial intelligence re‐construction was performed(AI3,AI6)simultaneously.A region of interest was drawn in four sets of images,and the signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)of the images were calculated.Different sequence images were qualitatively evaluated from four aspects:gray-white matter contrast,deep nucleus contour,tongue muscle noise,and optic nerve clarity.The SNR and CNR differences of the images were analyzed by single factor repeated measurement analysis,and the Friedman test was used to compare the qualitative analysis results.Results Under the same CS fac‐tors,the SNR and CNR of AI-reconstructed images were significantly better than conventional CS-reconstructed images(AI3_(SNR),20.63±6.90,vs CS3_(SNR),17.12±5.27,P<0.001;AI3_(CNR),12.93±4.79 vs CS3_(CNR)10.94±3.70,P<0.001;AI6_(SNR),20.60±7.46,vs CS6_(SNR),12.33±3.22,P<0.001;AI6_(CNR)12.84±5.20 vs CS6_(CNR)7.66±2.43,P<0.001).In the qualitative analysis of images,the scores of AI re‐constructed images were significantly higher than those of convention‐al reconstructed images(AI3,11.56±0.72 vs CS3,11.29±0.75,P<0.001;AI6,11.67±0.61 vs CS6,7.51±0.84,P<0.001).The SNR and CNR of AI6 images were higher than those of CS3 images,how‐ever,there was no significant statistical difference.There was no significant difference between the overall score values of CS3 and AI6 images,but it was significantly higher than that of CS6 images(P<0.001).Conclusion Compared with CS3,AI6 can shorten the scanning time by nearly half on the basis of ensuring the resolution of the 3D T1WI.