ABSTRACT
Objective@#To detect the feasibility and efficiency of bone age(BA) artificial intelligence(AI) estimation based on deep learning features from traditional regions of interest(ROI) in hand digital radiographs(DR).@*Methods@#BA dataset of left hand DR with 11 858 subjects aged from 0 to 18 years in Children′s Hospital of Shanghai were split to training(80.0%) and validation (20.0%) set in this study. An improved regression convolutional neural networks and extreme gradient boosting decision tree method were utilized for the BA analysis based on traditional ROIs in the images. Another set of BA data with 1 229 subjects also in the hospital was adopted for test. Mean average precision(mAP) and mean absolute error(MAE) were used to assess model accuracy of detection and BA prediction, respectively.@*Results@#The mAP of ROIs detection of the model was 0.91,and MAE of all male and female subjects was 0.461 and 0.431 years respectively in validation and test sets. The difference less than 1 year in test accounted for 90.07% between BA assessment of the model and of the peadiatric radiologists, with an accuracy rate of 96.67%.The difference over 1 year was 9.03% (with underestimation of 6.43% and overestimation of 2.60%), in which corresponding age data was of being less in training set or sesamoid nearby adductor pollicis or fusion of epiphysis appeared in test set.@*Conclusion@#An AI model based on deep learning of traditional ROIs′ features in hand DR images is initially achieved to automatically predict BA rapidly and effectively, yet it still needs further optimization.
ABSTRACT
s] Objective To detect the feasibility and efficiency of bone age(BA) artificial intelligence(AI) estimation based on deep learning features from traditional regions of interest(ROI) in hand digital radiographs(DR). Methods BA dataset of left hand DR with 11 858 subjects aged from 0 to 18 years in Children′s Hospital of Shanghai were split to training(80.0%) and validation (20.0%) set in this study. An improved regression convolutional neural networks and extreme gradient boosting decision tree method were utilized for the BA analysis based on traditional ROIs in the images. Another set of BA data with 1 229 subjects also in the hospital was adopted for test. Mean average precision(mAP) and mean absolute error(MAE) were used to assess model accuracy of detection and BA prediction, respectively. Results The mAP of ROIs detection of the model was 0.91,and MAE of all male and female subjects was 0.461 and 0.431 years respectively in validation and test sets. The difference less than 1 year in test accounted for 90.07% between BA assessment of the model and of the peadiatric radiologists, with an accuracy rate of 96.67%.The difference over 1 year was 9.03% (with underestimation of 6.43% and overestimation of 2.60%), in which corresponding age data was of being less in training set or sesamoid nearby adductor pollicis or fusion of epiphysis appeared in test set. Conclusion An AI model based on deep learning of traditional ROIs′features in hand DR images is initially achieved to automatically predict BA rapidly and effectively, yet it still needs further optimization.
ABSTRACT
OBJECTIVE： To establish the method for the rapidly non-destructive quality control of Liuwei dihuang capsule. METHODS： AOTF-NIR spectrometry was adopted. Taking 80 batches of Liuwei dihuang capsule produced by a manufacturer in recent three years as samples， HPLC chromatogram was adopted to determine the contents of loganin， morroniside， paeonol， paeoniflorin and ursolic acid； the content of water was determined according to general principles stated in 2015 edition of Chinese Pharmacopeia （part Ⅰ）. Taking 70 batches of samples as correction set， the partial least square method and the cross-validation algorithm were used to establish the NIR quantitative model of 6 indexes in Liuwei dihuang capsules with the Unscrambler quantitative analysis software. Taking residual 10 batches of samples as validation set， external validation was conducted for the model. RESULTS： The correlation coefficients （R2） of internal and external validation of loganin， morroniside， paeonol， paeoniflorin， the content of water quantitative model were all greater than 0.9； the correction of standand deviation （RMSEC） were 0.372 8， 0.025 4， 0.263 3， 0.288 5， 0.186 7 and 0.037 7； the prediction of standard deviation （RMSEP） were 0.462 2， 0.077 5， 0.472 1， 0.634 9， 0.293 4 and 0.206 9； the external verification showed that mean deviations of preclicted value to actual value were 6.04％， 6.05％， 5.87％， 6.97％， 5.62％ and 4.83％， with the mean deviation less than 10％.CONCLUSIONS：The established method can achieve rapidly non-destructive analysis Liuwei dihuang capsule.
ABSTRACT
OBJECTIVE:To establish the method for rapid judgement of blending endpoint of Jingqi shuangshen capsules and content determination of astragaloside Ⅳ. METHODS:AOTF-NIR combined with principal component analysis and Moving Block Standard Deviation method was used to identify the blending endpoint. First derivative combined with savitzky-golay filter method were used to spectrum pretreatment. The partial least square method was used to establish quantitative analysis model of the content of astragaloside Ⅳin mixed endpoint sample. The content of astragaloside Ⅳ in mixed endpoint sample was determined by HPLC-ELSD to validate the model. RESULTS:Methodology validation of content determination of astragaloside Ⅳ in mixed material sample and mixed endpoint sample was in line with the requirements. NIR monitoring results showed that the product reached the blending endpoint after 30 min. The results of NIR monitoring were generally consistent with the results of HPLC-ELSD. The principal component dimension of the quantitative model was 9;determination coefficients was 0.954 9;Root Mean Square of Calibration of the model was 0.039 2;Root Mean Square Error of Prediction of the model was 0.042 6. Predicted average value of astragaloside Ⅳ by NIR was 11.74 mg/g,and measured average value of astragaloside Ⅳ by HPLC-ELSD was 11.38 mg/g;average deviation was 3.16%. CONCLUSIONS:AOTF-NIR can rapidly judge the blending endpoint sample of Jingqi shuangshen capsules,rapidly determine the content of astragalosideⅣin mixed endpoint material,improve the quality control level of blending process and shorten blending cycle.
ABSTRACT
OBJECTIVE:To establish rapid method for content determination of cryptotanshinone in Salvia miltiorrhiza. METHODS:The content of cryptotanshinone in sample was determined by HPLC(as reference value). AOTF-NIDRS combined with PLS was used to establish quantitative correction model for the content of cryptotanshinone in S. miltiorrhiza. According to the results of content determination of cryptotanshinone in samples,35 samples of medicinal material were collected. First-order derivative combined with smoothing filter coefficient method was used to pretreat spectrum,and optimal band range for content determination of cryptotanshinone in sample ranged 1 250-2 150 nm. RESULTS:Methodology validation of content determination of cryptotanshinone in sample was in line with the requirements. Correction mean square deviation of quantitative correction model of cryptotanshinone was 0.014 6,and predicted mean square deviation was 0.022 3,coefficient of association was 0.976 6. The internal verification deviation was 2.41% and the external verification deviation was 4.06%. CONCLUSIONS:This method is rapid,accurate,simple and pollution-free.It can be used for rapid content determination of cryptotanshinone in S.miltiorrhiza.
ABSTRACT
OBJECTIVE:To establish the method for rapid quality evaluation of Astragali Radix.METHODS:The moisture of medicinal material was determined by oven drying method;the content of astragaloside Ⅳ was determined by HPLC-ELSD;the content of isoflavone glucoside was determined by HPLC (as reference value).The partial least squares (PLS) method combined with acousto-optic turnable filter-NIDRS was adopted to build quantitative model of above indexes in Astragali Radix (as predict value).According to reference value,60 batches of sample were collected.The spectra pretreatment was conducted by first derivative method combined with Savitzky golay.The optimal bands of moisture,astragaloside Ⅳ and isoflavone glucoside were 1 100-2 300 nm,1 080-2 160 nm,1 170-2 230 nm,respectively.RESULTS:The content determination of moisture,astragaloside Ⅳ and isoflavone glucoside in samples were all in line with methodology requirements.The corrected mean square root deviation of quantitative model for moisture,astragaloside Ⅳ,calycosin glucoside were 0.132 3,0.006 6,0.002 5,respectively;predicted mean square root deviation were 0.237 1,0.016 3,0.004 7;internal cross validation coefficient of correction set were 0.975 9,0.953 3,0.968 0;internal verification deviation of quantitative model were 1.43％,1.90％,1.84％;external verification deviation were 1.73 ％,2.68 ％,2.71％,respectively.CONCLUSIONS:The method is rapid,accurate,simple,pollution-free,and can be used for rapid quality evaluation of Astragali Radix.