Link to bioRxiv paper:
http://biorxiv.org/cgi/content/short/2023.01.24.525271v1?rss=1
Authors: Kaderuppan, S. S., Wong, E. W. L., Sharma, A., Woo, W. L.
Abstract:
We present herein a triplet string of concatenated O-Net ( bead) architectures (formulated as discussed in our previous study) which we term {Theta}-Net as a means of improving the viability of generated super-resolved (SR) images in silico. In the present study, we assess the quality of the afore-mentioned SR images with that obtained via other popular frameworks (such as ANNA-PALM, BSRGAN and 3D RCAN). Models developed from our proposed framework result in images which more closely approach the gold standard of the SEM-verified test sample as a means of resolution enhancement for optical microscopical imaging, unlike previous DNNs. In addition, cross-domain (transfer) learning was also utilized to enhance the capabilities of models trained on DIC datasets, where phasic variations are not as prominently manifested as amplitude/intensity differences in the individual pixels [unlike phase contrast microscopy (PCM)]. The present study thus demonstrates the viability of our current multi-paradigm architecture in attaining ultra-resolved images under poor signal-to-noise ratios, while eliminating the need for a priori PSF & OTF information. Due to the wide-scale use of optical microscopy for inspection & quality analysis in various industry sectors, the findings of this study would be anticipated to exhibit a far-ranging impact on several engineering fronts.
Copy rights belong to original authors. Visit the link for more info
Podcast created by Paper Player, LLC
