Commentary on "Application of Photoshop-based Image Analysis to Quantification of Hormone Receptor Expression in Breast Cancer".

Abstract

At the time of publication of this article, image analysis as applied to diagnostic surgical pathology was a nascent technology, often characterized as a “solution in search of a problem.” After all, biomarkers such as estrogen receptor (ER) in breast cancer were scored in a threshold manner, with the assumption that this scoring was highly accurate and reproducible. Unfortunately, several studies in the early 2000s found significant discordance in this ER evaluation when slides were re-reviewed.1,2 Ironically, prior approaches to determining ER expression in breast cancer, for example, the dextran-coated charcoal method, were quantifiable and reproducible, but required fresh frozen tissue and could not be used on formalin-fixed paraffin-embedded tissue. Image analysis on deparaffinized, formalin-fixed tissue sections immunostained with appropriate antibodies was anticipated to be a method that could increase the predictive power of IHC determination of hormone receptor status. Indeed, this 1997 paper was one of the first to employ image analysis in the evaluation of ER. While after more than two decades Photoshop has maintained its role as a premier raster graphics and digital art editor, it has not taken a second life as an image analysis tool. We selected Photoshop as a “maverick” technique that permitted us to accomplish the analysis we desired, but to also avoid the use of commercial image analysis instruments and software, which at that time were expensive and unaffordable for our laboratory. Today, while commercial image analysis hardware and software are legion, free public domain software such as QuPath can yield comparable results as many commercial software packages3 and far exceed the capabilities of Photoshop. While our 1997 paper succeeded as a “proof of principle,” now 25 years later, ER is still not quantified in the vast majority of cases, but is instead still “eyeballed” to determine whether more than 1% of tumor cells are positive for nuclear signal (or between 1% and 10%), but no image analysis–based quantification is required.4 There may still be a role in the future for more quantitative ER determination in breast cancer, but that will be determined not by technology alone but largely by evidence that may be forthcoming from clinical studies. But it will almost certainly not involve the use of Photoshop!