Lightweight Deep Learning Network Assisted Cell Classification Using Lensless Computational Microscopic Imaging Data

Zhaohui Wang; Huan Kang; Duofang Chen; Xinyi Xu; Qi Zeng; Jimin Liang; Xueli Chen

doi:10.3788/CJL202249.0507204

Chinese Journal of Lasers, Volume. 49, Issue 5, 0507204(2022)

Lightweight Deep Learning Network Assisted Cell Classification Using Lensless Computational Microscopic Imaging Data

Zhaohui Wang¹, Huan Kang¹, Duofang Chen¹, Xinyi Xu¹, Qi Zeng¹, Jimin Liang^2、**, and Xueli Chen^1、*

¹Xi’an Key Laboratory of Intelligent Sensing and Regulation of Trans-Scale Life Information, School of Life Science and Technology, Xidian University, Xi’an 710126, Shaanxi, China

²School of Electronic Engineering, Xidian University, Xi’an 710126, Shaanxi, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(8)

Fig. 1. Lens-less computational microscopic imaging and classification system

Download full size

Fig. 2. Microscopic images of ECa109 cells under lensless computational microscopic imaging system and 4× objective lens. (a) Full field-of-view image of ECa109 cells acquired by lensless computational microscopic imaging system; (b) microscopic image of ECa109 cells in square outlined in Fig. 2(a) under 4× objective lens; (c) enlarged image of ECa109 cells in square outlined in Fig. 2(a) acquired by lensless computational microscopic imaging system

Download full size

Fig. 3. Images of different cells. (a) SUM cell; (b) MCF10A cell; (c) ECa109 cell; (d) CL-1 cell

Download full size

Fig. 4. Network structure of Depthwise-ResNeXt

Download full size

Fig. 5. Fuzzy matrix of Depthwise-ResNeXt

Download full size

Table 1. Experimental results of networks with different sizes

View table

Table 1. Experimental results of networks with different sizes

Way to improve	Channel number					Test accuracy
Way to improve	Conv 1	Layer 1	Layer 2	Layer 3	Layer 4	Test accuracy
Depthwise convolution	32	32	64	128	256	0.889
Depthwise convolution	64	64	128	256	512	0.928
Depthwise convolution	64	64	128	256	0	0.905
Depthwise convolution	64	128	256	512	1024	0.926

Table 2. Experimental results of different structured networks

View table

Table 2. Experimental results of different structured networks

Index	Channel shuffle	Depthwise convolution	Asymmetric convolution	Depthwise convolution & channel shuffle	Depthwise convolution & asymmetric convolution	All	ResNeXtM
Accuracy	0.922	0.928	0.922	0.913	0.911	0.908	0.921
Calculated amount /10⁹	1.04	1.01	1.03	1.01	1.01	1.00	1.04
Number of parameters /kB	846	806	830	806	803	803	846
Time /s	1.17±5×10^－4	1.12±4.2×10^－4	1.18±1.4×10^－4	1.13±2.2×10⁵	1.12±2.4×10⁵	1.15±2.2×10^－4	1.16±6.4×10^－5

Table 3. Comparison of different network structures

View table

Table 3. Comparison of different network structures

Index	ShuffleNet	MobileNet	ResNet	DenseNet	GoogleNet	Depthwise-ResNeXt
Accuracy	0.885	0.870	0.924	0.931	0.924	0.928
Calculated amount/10⁹	253.406×10^－3	1.944	10.460	10.247	4.88	1.01
Number of parameters /kB	345.460	2.228×10³	11.172×10³	6.952×10³	5.598×10³	806
Run time /s	1.25±26×10^－4	1.33±4.8×10^－4	1.19±1.6 ×10^－5	1.88±9.4×10^－4	1.33±2.1×10^－4	1.12±4.2×10^－4

Tools

Get Citation

Copy Citation Text

Zhaohui Wang, Huan Kang, Duofang Chen, Xinyi Xu, Qi Zeng, Jimin Liang, Xueli Chen. Lightweight Deep Learning Network Assisted Cell Classification Using Lensless Computational Microscopic Imaging Data[J]. Chinese Journal of Lasers, 2022, 49(5): 0507204

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Received: Oct. 15, 2021

Accepted: Jan. 10, 2022

Published Online: Mar. 9, 2022

The Author Email: Liang Jimin (jimleung@mail.xidian.edu.cn), Chen Xueli (xlchen@xidian.edu.cn)

DOI:10.3788/CJL202249.0507204

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology