Channel-Wise Attention Mechanism Relevant UNet-Based Diffraction-Limited Fluorescence Spot Detection and Localization

Yongjian Yu; Yue Wang; Huan Li; Wenchao Zhou; Fengfeng Shu; Ming Gao; Yihui Wu

doi:10.3788/LOP230718

Laser & Optoelectronics Progress, Volume. 60, Issue 14, 1412004(2023)

Channel-Wise Attention Mechanism Relevant UNet-Based Diffraction-Limited Fluorescence Spot Detection and Localization

Yongjian Yu¹, Yue Wang^2、*, Huan Li², Wenchao Zhou², Fengfeng Shu², Ming Gao^2,3, and Yihui Wu^1,2、**

Author Affiliations

¹School of Ophthalmology & Optometry, Wenzhou Medical University, Wenzhou 325035, Zhejiang, China

²Key Laboratory of Optical System Advanced Manufacturing Technology, Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, Jilin, China

³University of Chinese Academy of Sciences, Beijing 100049, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. Overall architecture of our network. (a) Training network; (b) network prediction process; (c) structure of the SE residual module

Download full size

Fig. 2. Algorithm performance under images with different sizes, densities, and SNRs. (a) F1 under different sizes; (b) F1 under different densities; (c) F1 under different SNRs; (d) localization error under different sizes; (e) localization error under different densities; (f) localization error under different SNRs

Download full size

$Prediction for high-density spot images. (a) Fluorescent spot prediction reaching the diffraction limit; (b) prediction based on density image; (c) detection results of each algorithm in the case of 1200 spots per image$

Fig. 3. Prediction for high-density spot images. (a) Fluorescent spot prediction reaching the diffraction limit; (b) prediction based on density image; (c) detection results of each algorithm in the case of 1200 spots per image

Download full size

Fig. 4. Representative images of datasets and their corresponding prediction

Download full size

Table 1. Performance of SE-Res-UNet using different density map loss functions
View table
Table 1. Performance of SE-Res-UNet using different density map loss functions
Loss F1 /% RMSE /pixel Recall /% Precision /%
Our loss 84.6 0.545±0.357 82.4 88.4
AWing loss 83.0 0.564±0.348 80.1 89.0
BCE loss 80.2 0.539±0.337 76.9 94.0

Table 2. Performance of different models on vesicle dataset

View table

Table 2. Performance of different models on vesicle dataset

Model	F1 /%	RMSE /pixel	Recall /%	Precision /%	Parameter quantity
SE-Res-UNet（16）	84.5	0.549±0.356	82.7	88.1	772539
SE-Res-UNet（8）	84.3	0.542±0.354	81.6	89.2	210271
Res-UNet（16）	83.7	0.543±0.369	83.2	84.8	743311
Res-UNet（8）	83.9	0.548±0.354	81.4	88.2	758531
SE-UNet（8）	84.0	0.557±0.351	81.1	89.3	141615
UNet（8）	83.4	0.570±0.351	81.9	86.1	138003

Table 3. Detection and localization performances of different algorithms under different SNRs

View table

Table 3. Detection and localization performances of different algorithms under different SNRs

SNR	SE-Res-UNet		deepBlink		Big-FISH
SNR	F1 /%	RMSE /pixel	F1 /%	RMSE /pixel	F1 /%	RMSE /pixel
1	31.2±13.3	1.023±0.134	33.1±11.6	1.115±0.076	9.6±3.4	1.036±0.148
2	90.2±8.1	0.620±0.075	88.2±8.1	0.645±0.083	66.4±16.1	0.736±0.108
3	98.9±0.6	0.362±0.023	97.8±1.0	0.385±0.031	93.1±2.1	0.423±0.033
4	99.2±0.6	0.264±0.028	98.5±0.9	0.292±0.039	96.9±2.1	0.350±0.069
7	99.3±0.6	0.151±0.017	98.7±0.7	0.188±0.038	98.1±1.4	0.271±0.075

Table 4. Detection and localization performances of different algorithms under different densities

View table

Table 4. Detection and localization performances of different algorithms under different densities

Density	SE-Res-UNet		deepBlink		Big-FISH
Density	F1 /%	RMSE /pixel	F1 /%	RMSE /pixel	F1 /%	RMSE /pixel
200	99.1±0.50	0.062±0.0041	97.0±0.98	0.296±0.0435	98.9±0.61	0.350±0.0715
400	99.0±0.30	0.066±0.0026	94.5±0.58	0.398±0.0193	97.3±1.10	0.409±0.0252
600	98.8±0.31	0.073±0.0037	91.9±0.70	0.502±0.0251	95.8±1.00	0.442±0.0185
800	98.6±0.35	0.081±0.0040	88.7±0.54	0.590±0.0209	95.0±1.50	0.455±0.0414
1000	98.3±0.24	0.096±0.0044	85.0±0.66	0.679±0.0152	77.7±7.79	0.592±0.0412
1200	97.6±0.25	0.115±0.0047	81.4±0.56	0.745±0.0104	68.3±11.81	0.702±0.0634

Table 5. Detection and localization performances of different algorithms on different datasets

View table

Table 5. Detection and localization performances of different algorithms on different datasets

Dataset	deepBlink		SE-Res-UNet		Big-FISH
Dataset	F1 /%	RMSE /pixel	F1 /%	RMSE /pixel	F1 /%	RMSE /pixel
SunTag	83.0	0.382±0.289	85.4	0.595±0.266	62.8	0.553±0.298
Vesicle	82.9	0.577±0.384	83.8	0.531±0.363	67.8	0.654±0.341
Receptor	80.6	0.512±0.320	80.7	0.471±0.305	72.8	0.519±0.272
Dense spots	90.6	0.510±0.167	98.6	0.079±0.017	90.1	0.478±0.118

Tools

Get Citation

Copy Citation Text

Yongjian Yu, Yue Wang, Huan Li, Wenchao Zhou, Fengfeng Shu, Ming Gao, Yihui Wu. Channel-Wise Attention Mechanism Relevant UNet-Based Diffraction-Limited Fluorescence Spot Detection and Localization[J]. Laser & Optoelectronics Progress, 2023, 60(14): 1412004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Instrumentation, Measurement and Metrology

Received: Feb. 27, 2023

Accepted: Apr. 17, 2023

Published Online: Jul. 14, 2023

The Author Email: Wang Yue (yihuiwu@ciomp.ac.cn), Wu Yihui (wangyue@ciomp.ac.cn)

DOI:10.3788/LOP230718

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Performance of SE-Res-UNet using different density map loss functions

Table 1. Performance of SE-Res-UNet using different density map loss functions

Table 2. Performance of different models on vesicle dataset

Table 2. Performance of different models on vesicle dataset

Table 3. Detection and localization performances of different algorithms under different SNRs

Table 3. Detection and localization performances of different algorithms under different SNRs

Table 4. Detection and localization performances of different algorithms under different densities

Table 4. Detection and localization performances of different algorithms under different densities

Table 5. Detection and localization performances of different algorithms on different datasets

Table 5. Detection and localization performances of different algorithms on different datasets