Advanced Imaging, Volume. 1, Issue 2, 021001(2024)
High-throughput, nondestructive, and low-cost histological imaging with deep-learning-assisted UV microscopy
Figures & Tables(7)
Fig. 1. Dual-modality tissue imaging systems for UV-LED and UV-LS image acquisition with the deep-learning-based contrast enhancement frameworks. (a) Overview of the dual-modality imaging system, facilitating both UV-LED and UV-LS imaging. Both modes share an optical path for data acquisition. (b) Detailed setup of the tissue scanning region, incorporating a quartz solid immersion lens to correct refraction in the illumination and detection paths. (c) Selection of a 2D image layer close to the sample surface from the 3D UV-LS image stack. (d) Application of the U-Frame network to enhance the contrast of UV-LED images by learning from UV-LS images.
Fig. 2. Comparison among UV-LED, CE-LED, UV-LS, and the corresponding H&E-stained images on thick healthy mouse brain tissue. (a) UV-LED image of half of the mouse brain; (b) CE-LED image generated from the UV-LED image in (a); (c) UV-LS image of the same mouse brain specimen as (a); (d) corresponding H&E-stained image obtained after imaging with UV-LED and UV-LS modes; (e) cross-sectional area measurement of (h)–(j); (f) intercellular distance measurement of (h)–(j); (g)–(j) zoomed-in views of green dashed rectangular regions marked in (a)–(d), respectively; (k)–(n) zoomed-in views of blue rectangular regions marked in (a)–(d), respectively.
Fig. 3. Comparison of contrast enhancement results using cycleGAN, pix2pix, and U-Frame compared to the original UV-LED and UV-LS on human lung cancer tissue. (a) UV-LED image. (b)–(d) CycleGAN, pix2pix, and U-Frame results of CE-LED images generated from (a), respectively; (e) UV-LS image of the same specimen as in (a); (f)–(j) zoomed-in views of yellow dashed rectangular regions marked in (a)–(e), respectively; (k)–(o) zoomed-in views of light green rectangular regions marked in (a)–(e), respectively; (p)–(t) zoomed-in views of red rectangular regions marked in (a)–(e), respectively.
Fig. 4. Comparison among UV-LED, CE-LED, UV-LS, and the corresponding H&E-stained images on thick human lung cancer tissue. (a) UV-LED image of lung cancer tissue; (b) CE-LED image generated from the UV-LED image in (a); (c) UV-LS image of the same specimen in (a); (d) corresponding H&E-stained image after acquiring (a) and (c); (e)–(h) zoomed-in views of red rectangular regions marked in (a)–(d), respectively; (i)–(l) zoomed-in views of blue dashed rectangular regions marked in (a)–(d), respectively; (m)–(p) and (q)–(t) are the further zoom-in images highlighted by the yellow dashed rectangular regions and the light-green rectangular regions marked in (i)–(l), respectively.
Fig. 5. Comparison of the virtual staining results of CE-LED and UV-LED. (a) CE-LED image of lung cancer tissue; (b) corresponding virtual staining result of (a) generated by the U-Frame virtual staining model; (c) corresponding H&E-stained image of the yellow rectangular region marked in (a); (d)–(e) zoomed-in views of the yellow rectangular regions marked in (a) and (b), respectively; (f) original UV-LED image of (d); (g) virtual staining result of (f).
Table 1. PSNR and FID Comparison among the CE-LED Images Generated by CycleGAN, Pix2pix, and U-Frame
View tableView in ArticleTable 1. PSNR and FID Comparison among the CE-LED Images Generated by CycleGAN, Pix2pix, and U-Frame
Comparison among the CE-LED Results Generated by Different Models PSNR FID CE-LED by cycleGAN [Fig. 3(b) ] to UV-LS [Fig.3(e) ]28.25 88.38 CE-LED by pix2pix [Fig. 3(c) ] to UV-LS [Fig.3(e) ]28.55 67.03 CE-LED by U-Frame [Fig. 3(d) ] to UV-LS [Fig.3(e) ]28.50 29.90
Table 2. PSNR and SSIM Comparison between UV-LED to UV-LS, and CE-LED to UV-LS in Human Lung Cancer Tissue
View tableView in ArticleTable 2. PSNR and SSIM Comparison between UV-LED to UV-LS, and CE-LED to UV-LS in Human Lung Cancer Tissue
Human Lung Cancer Tissue Image PSNR SSIM UV-LED [Fig. 4(i) ] to UV-LS [Fig.4(k) ]27.49 0.72 CE-LED [Fig. 4(j) ] to UV-LS [Fig.4(k) ]29.04 0.78
Tools
Get Citation
Copy Citation Text
Jiajie Wu, Weixing Dai, Claudia T. K. Lo, Lauren W. K. Tsui, Terence T. W. Wong, "High-throughput, nondestructive, and low-cost histological imaging with deep-learning-assisted UV microscopy," Adv. Imaging 1, 021001 (2024)
Paper Information
Category: Research Article
Received: May. 22, 2024
Accepted: Jul. 1, 2024
Published Online: Aug. 8, 2024
The Author Email: Terence T. W. Wong (ttwwong@ust.hk)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20