A Clearer Picture
How does a new quality control tool filter digital slide images by quality?
Luke Turner | | Quick Read
Knife chatter – a term with which some pathologists will be all too familiar – refers to a compromise in the quality of glass slides caused by issues with slide preparation, such as air bubbles, smears, or ragged cuts in the tissue. Pathologists analyzing slides under a microscope can easily identify those affected by the issue – but when it comes to digital pathology imaging, knife chatter introduces a whole new standardization problem. To address it, Anant Madabhushi and Andrew Janowczyk of Case Western Reserve University’s Center for Computational Imaging and Personal Diagnostics have developed a program that aims to ensure the quality of digital slide images.
As digital pathology continues to alter the landscape of clinical diagnostic workflows, more and more physicians have turned to digital imaging systems to analyze tissue. At the moment, though, there are no standards for the preparation and digitization of slides – and practically perfect ones are routinely found alongside those of poor quality. In the context of machine learning, this can mislead computer programs trying to recognize the appearance of cancerous cells.
When Janowczyk discovered that about 10 percent of the 800 cancer samples he reviewed in The Cancer Genome Atlas (TCGA) had issues, such as cracked slides or air bubbles, he decided to create an application to help. The new quality control tool, called HistoQC, prevents the need to manually review glass and digital slides, instead offering an automated approach.
HistoQC locates artifacts in slides that need to be reproduced and identifies areas unsuitable for computational analysis. The program uses a combination of image metrics, such as brightness and contrast, alongside features such as edge detectors and supervised classifiers to pinpoint the slide regions that are most accurate. Users can then monitor and filter the slides in real-time and explicitly define acceptable artifact tolerances. When two pathologists reviewed HistoQC on 450 slides from TCGA, the output was suitable for computational analysis in over 95 percent of cases (1).
Having recently secured a three-year, US$1.2 million grant from the National Cancer Institute to further develop HistoQC, Madabhushi describes the technology as a step toward the “democratization of imaging technology.” The team hopes to accelerate the widespread use of AI for interrogating tissue images and, to this end, they have made HistoQC an open-source platform – freely available for all to access, modify, and extend via an online repository.
- A Janowczyk et al., “HistoQC: An open-source quality control tool for digital pathology slides”, JCO Clin Cancer Inform, 3, 1–7 (2019). PMID: 30990737.