Image Analysis Quality Control Part 1: Image Scan QC
14 May 2020
By Hannah Thomson

Good quality image analysis requires good quality images. Therefore, before beginning quantitative analysis it is essential to first conduct a Quality Control (QC) review of all images to assess suitability for analysis and to only proceed with images that meet certain criteria. At OracleBio, we QC check every image before analysis, taking into consideration 1) Image Scan 2) Tissue Section and 3) Staining. In the first of our 3-part Image Analysis Quality Control Blog series, we look at a number of examples of pitfalls in Image Scan QC and the impact they can have on image analysis data if not identified and remedied.

Magnification and Resolution:

Magnification is the degree to which the image is made bigger, to make viewing easier.  A more important feature, the resolution, is the ability to discern features of a certain distance apart e.g. one can distinguish features that are 1um apart for example, or 10um apart.  Resolution of an image seen on a computer monitor will be governed by the initial magnification of the image (defined by the objective used on the scanner), the number and size of the pixels of the scanner camera and finally, for visual annotations and checking, by the number and size of pixels on the viewing monitor.

For example, OracleBio’s scanner, the Hamamatsu Nanozoomer HT2.0, will capture a 20X image onto the camera with a resolution of 0.46 µm per pixel, equivalent to 2.17 pixels per µm.  With the size of a mammalian cell ranging between 10-100µm, one can see why x20 magnification is sufficient for the assessment of many of the biological questions assessed by image analysis, such as cell number, cell-cell and cell to border spatial proximity.  However, 40X magnification may be required if distinguishing between membrane and cytoplasmic staining or quantifying RNA staining.


Areas of tissue that are ‘out-of-focus’ (OoF) can create problems during the image analysis process, for example by interfering with the correct classification of the affected tissue area and potentially preventing accurate detection of cells during algorithm development.  Avoiding this confounding factor should be approached at broadly three levels;

1) Global OoF, whereby a scanner error has condemned the whole image to waste and repeat of the scanning process is necessary

2) Regional OoF, that affects only part of a slide, maybe caused by slide folds in the tissue section or again a scanner error that only effects some tiles that make up the whole image.  If significant these areas can be negatively annotated and removed from the analysis.

3) Local OoF, whereby small areas of the slide, often difficult to discern, are affected.  Again, a judgement should be made about the extent of this and its impact on the subsequent analysis. A deep learning algorithm to identify and exclude any OoF areas could potentially be  developed to alleviate these issues.

Right: An example of regional OoF. Two sections from the same image, one out of focus (top) and one in focus (bottom).

Scan Lines:

Scan lines is a generic term for a number of image aberrations caused by the scanning process.  Whole slide images are compiled into a single file from an array of individual scanner runs (tiles) that are electronically stitched together.  Scan lines are seen when:

  • image tiles are not fully aligned.
  • Individual tiles may be OoF.
  • poor quality illumination, caused by a faulty scanner lamp, results in the appearance of banding in the background of images.

The presence of any scanner lines should be noted during the Image QC process and may also need to be manually excluded during annotation procedures.

Left: A section of an image scan with a number of scan lines present. 

Scanning Background:

The brightness of “clear glass” on a scan is a clue to the illumination level used for the scanning process.  The degree of illumination obviously affects the brightness of the image and this is linked to the threshold levels that are set in the image analysis software to distinguish between features.  Ideally this illumination level should be kept constant for slides images that are to be analysed and results compared.  However, when images are  run in batches that are collected over many weeks/months, as can happen for clinical trials, then this standardisation may be lost; changes to the scanner setting if SOPs are not followed, use of different scanners and ageing of scanner lamps are some of the possible reasons this may occur.  Of course, if not too severe then this variation can be corrected by adjustments to the App settings for the image analysis.  But best to avoid this, if possible, and certainly for those clinical studies performed under a quality framework (GCP).

Right: Two sections from the same image showing inconsistent scanning background.

Immunofluorescence Bleed through:

If the fluorophores used in immunofluorescence studies are too close together in wavelength, then obviously there is the chance of the light from one channel spilling over into another, termed spectral overlap. This would have a negative impact on the analysis when trying to identify specific staining for the IF markers.

Right: Immunofluorescence bleed through. The 610 nm, 570 nm and 767 nm channels represent membrane/cytoplasmic staining. However, fluorescent overlap from the 670 nm channel results in cells being quantified in the nuclear compartment giving rise to false positive cell detection due to this bleed through effect.

There are many factors to consider when deciding if an image is suitable for image analysis and these are just some to include when conducting Image Scan QC. To find out what else we include in our QC check, click here to read Image Analysis Quality Control Part 2: Tissue Section QC and Image Analysis Quality Control Part 3: Staining QC.

Encountering some of the issues above does not necessarily mean that an image cannot be analysed as some problems can be overcome, for example by negative annotation or additional algorithm development. Unfortunately, at OracleBio we do have to fail a significant number of images every year because the image analysis data generated will not be robust. If you are unsure if your images are suitable for quantitative analysis or need advice on how to avoid encountering some of these issues, please contact us. 

About the Author: Nicole Stillie

Senior Image Analysis Scientist

Nicole joined OracleBio in 2016 after graduating with a B.Sc. in Biomedical Science. She is now a Senior Image Analysis Scientist and QCs hundreds of images every month. Connect with Nicole on LinkedIn. 

Discover how we maintain high levels of quality

Explore OracleBio's Image Analysis Blog

Stay up to date with OracleBio

Subscribe to our mailing list to stay informed about news, webinars, events and more