Identifying Multi-Class Drugs by Using Near-Infrared Spectroscopy and Variational Auto-Encoding Modeling

An-bing ZHENG; Hui-hua YANG; Xi-peng PAN; Li-hui YIN; Yan-chun FENG

doi:10.3964/j.issn.1000-0593(2020)12-3946-07

Journals >Spectroscopy and Spectral Analysis >Volume 40 >Issue 12 >Page 3946 > Article

Spectroscopy and Spectral Analysis
Vol. 40, Issue 12, 3946 (2020)

Identifying Multi-Class Drugs by Using Near-Infrared Spectroscopy and Variational Auto-Encoding Modeling

An-bing ZHENG^1、1、*, Hui-hua YANG^1、1, Xi-peng PAN^1、1, Li-hui YIN^1、1, and Yan-chun FENG^1、1

Author Affiliations

¹[in Chinese]

¹1. School of Automation, Beijing University of Posts and Telecommunications, Beijing 100876, China

show less

DOI: 10.3964/j.issn.1000-0593(2020)12-3946-07 Cite this Article

An-bing ZHENG, Hui-hua YANG, Xi-peng PAN, Li-hui YIN, Yan-chun FENG. Identifying Multi-Class Drugs by Using Near-Infrared Spectroscopy and Variational Auto-Encoding Modeling[J]. Spectroscopy and Spectral Analysis, 2020, 40(12): 3946 Copy Citation Text

EndNote(RIS)

BibTex

Plain Text

show less

Fig. 1. Spectra of four kinds of drugs with a large number of samples, peak and valley positions overlap mostly

Download full size | View in the Article

Similar Spectra of the same drug produced by different manufactures. Metformin hydrochloride tablets came from No.6 and 7. Chlorphenamine maleate tablets came from No.18 and 19

Fig. 2. Similar Spectra of the same drug produced by different manufactures. Metformin hydrochloride tablets came from No.6 and 7. Chlorphenamine maleate tablets came from No.18 and 19

Download full size | View in the Article

Fig. 3. Overall structure block diagram

Download full size | View in the Article

Drug Name	Product (manufacturer) counts	Spectra samples
metformin hydrochloride tablets	14	94,48,67,21,48,64,27,35,48,24,68,97,97,97
chlorpromazine hydrochloride tablets	5	58,135,49,39,45
chlorphenamine maleate tablets	5	39,45,36,39,94
cefuroxime axetil tablets	5	56,29,27,89,37
Total		1 721

Table 1. Experimental data of NIR

View in the Article

Train/Test	Precise	Recall	F1	Accuracy
1 549/172	0.997	0.989	0.991	0.995
1 377/344	0.994	0.989	0.991	0.994
1 205/516	0.986	0.986	0.986	0.991
1 033/688	0.995	0.991	0.993	0.994
861/861	0.994	0.992	0.993	0.995
688/1 033	0.908	0.894	0.887	0.924
516/1205	0.893	0.861	0.861	0.905
344/1 377	0.816	0.782	0.782	0.853

Table 2. Experimental results under different training and test set partitions

View in the Article

Train/Test	VAE	PLS-DA	Linear SVM	RBF_SVM	K-NN	BP-ANN	DBN	SAE	CNN
1 549/172	0.997	0.967	0.962	0.945	0.906	0.929	0.940	0.979	0.995
1 377/344	0.994	0.952	0.946	0.955	0.879	0.918	0.946	0.971	0.980
1 205/516	0.986	0.966	0.948	0.931	0.904	0.927	0.923	0.976	0.979
1 033/688	0.995	0.961	0.937	0.934	0.887	0.883	0.936	0.960	0.964
861/861	0.994	0.961	0.930	0.916	0.861	0.892	0.925	0.955	0.949
688/1 033	0.908	0.959	0.929	0.893	0.825	0.890	0.895	0.920	0.963
516/1 205	0.893	0.967	0.897	0.863	0.796	0.881	0.897	0.917	0.899
344/1 377	0.816	0.957	0.849	0.826	0.748	0.845	0.853	0.911	0.845

Table 3. Accuracy of various multi-class classification algorithms

View in the Article

Train/Test	VAE	PLS-DA	Linear SVM	RBF_SVM	k-NN	BP-ANN	DBN	SAE	CNN
1 549/172	17.324/0.020	0.585/0.004	0.933/0.271	3.793/0.330	0.080/0.142	18.211/0.002	231.729/0.015	9.843/0.027	23.790/0.005
1 377/344	14.082/0.029	0.522/0.008	0.787/0.476	3.183/0.595	0.066/0.267	16.779/0.003	204.728/0.029	10.319/0.044	21.368/0.007
1 205/516	21.433/0.041	0.425/0.011	0.668/0.674	2.700/0.856	0.050/0.369	13.352/0.004	184.345/0.038	9.536/0.061	21.066/0.011
1 033/688	21.210/0.062	0.340/0.014	0.583/0.791	2.217/0.998	0.041/0.458	8.924/0.006	159.686/0.046	9.967/0.094	21.385/0.012
861/861	21.552/0.067	0.263/0.017	0.433/0.886	1.760/1.052	0.027/0.622	13.784/0.006	140.062/0.066	9.960/0.122	22.253/0.015
688/1 033	18.496/0.087	0.221/0.022	0.318/0.901	1.186/1.090	0.020/0.604	19.103/0.008	119.054/0.086	8.898/0.136	42.056/0.019
516/1 205	19.992/0.100	0.180/0.026	0.218/0.821	0.862/0.983	0.014/0.681	12.730/0.010	93.467/0.090	10.028/0.151	36.602/0.023
344/1 377	18.064/0.117	0.125/0.030	0.123/0.724	0.427/0.827	0.008/0.563	12.947/0.010	73.280/0.184	10.985/0.189	43.297/0.022

Table 4. Training and inferencing time of each algorithm (in second, training time/inferencing time)

Download Citation

Save the article for my favorites

Paper Information