# How to use SQUIRREL analysis as a performance evaluation metric?

**URL:** <https://forum.microlist.org/t/how-to-use-squirrel-analysis-as-a-performance-evaluation-metric/3279>\
**Category:** Uncategorized\
**Created:** [November 26, 2025, 5:15pm UTC](https://forum.microlist.org/t/how-to-use-squirrel-analysis-as-a-performance-evaluation-metric/3279 "2025-11-26T17:15:53Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Qiqi-Lu](https://avatars.discourse-cdn.com/v4/letter/q/a6a055/32.png) [@Qiqi-Lu](https://forum.microlist.org/u/Qiqi-Lu)\
**Post date:** [November 26, 2025, 5:15pm UTC](https://forum.microlist.org/t/how-to-use-squirrel-analysis-as-a-performance-evaluation-metric/3279/1 "2025-11-26T17:15:53Z")

</div>

Hi everyone,

I want to use RSE and RSP in SQUIRREL analysis as two metrics to evaluate the restoration performance of a deep learning network.

Now, I have a WF image (as network input), restored image (output of network), and SIM image (as ground-truth).

Can I use SIM image as the reference image for the SQUIRREL analysis, and calculate RSE and RSP of WF image and restored image respectively? (to evaluate the quality improvement from WF to restored image).

I saw that, in the paper of SQUIRREL analysis, a diffraction-limited image or a image with lower-resolution was used as reference. But in the above situation, a SIM image is used as reference which has a higher resolution then the WF image (which as used to calculate RSE and RSP). Will this cause any problems (especially when estimating resolution scaling function (RSF))?

 ![image](https://us1.discourse-cdn.com/flex020/uploads/microlist/original/2X/7/72b2b11924deffe48304000650574125bf7ee467.png)

In some studies, they use ground-truth image as reference image:

1. Weigert, M. _et al._ Content-aware image restoration: pushing the limits of fluorescence microscopy. _Nat Methods_ **15** , 1090–1097 (2018).
2. Fang, L. _et al._ Deep learning-based point-scanning super-resolution imaging. _Nat Methods_ **18** , 406–416 (2021).

While in some studies, they use a diffraction-limited image as reference image, even there exists ground-truth image.

1. Chen, J. _et al._ Deep-learning accelerated super-resolution radial fluctuations (SRRF) enables real-time live cell imaging. _Optics and Lasers in Engineering_ **172** , 107840 (2024).
2. Jin, L. _et al._ Deep learning enables structured illumination microscopy with low light levels and enhanced speed. _Nat Commun_ **11** , 1934 (2020).
3. Wang, H. _et al._ Deep learning enables cross-modality super-resolution in fluorescence microscopy. _Nat Methods_ **16** , 103–110 (2019). (Supplementary Note 7)

Should I use the WF image as reference and observe the consistence of the value of RSE between restored image and ground-truth image, or use the ground-truth image as reference and compare the value of RSE of restored image and WF image?

Looking forward to your reply!

(have a same topic in image.sc [[How to use SQUIRREL analysis as a performance evaluation metric? - Image Analysis - Image.sc Forum](https://forum.image.sc/t/how-to-use-squirrel-analysis-as-a-performance-evaluation-metric/117866?u=qiqi-lu%5C)])