Eliminating Subjectivity in Pathology
Eliminating Subjectivity in Pathology
Complete the form below and we will email you a PDF version of "Eliminating Subjectivity in Pathology"
Agilent recently released an online version of the Dako Atlas of Stains as a reference tool on the use and interpretation of stains. The new interactive version of the Atlas enhances its usefulness as a reference and extends its use into education and training of pathologists. Trainees can view and annotate the slide images, and the tool can assess whether the trainee is judging the stains accurately or not. The tool also serves as a reference guide for experienced pathologists, providing a framework for consistent assessment.
All this is necessary because one any of a million different factors can influence the intensity of a stain, its coloration, its bonding to the cells, and its appearance under the microscope. With training and experience, pathologists learn to adjust to these changes, but stain variability is one small example of the subjectivity that is endemic to pathology today. And, while subjectivity isn’t necessarily related to accuracy, the more subjective an observation, the less likely it is to be accurate.
Pathologists don’t even talk about accuracy as much as concordance – how much agreement there is between the diagnoses provided by the pathologists examining a particular slide. Give the same surgical tissue slide to 100 pathologists, for example, and as few as 75 of them will agree on a diagnosis. Concordance on breast tissue samples, on the other hand, runs between 90 and 99 percent.
Here are some of the many factors that make pathology such a subjective medical science:
Variability in staining. Because staining is a chemical process, it is subject to the vagaries involved in chemistry – contaminants in the water, variations in temperature, differences in chemical composition between batches of chemicals, speed of agitation, or a difference in protocol. All these can result in a difference in the appearance of stains. Stains from different laboratories, from different people in the same laboratory, or even different batches handled by the same person will present differently. Laboratories deal with this by including reference slides – with known characteristics – in every batch to provide a reference point for pathologists when interpreting results. Pathologists also learn to compensate for variations when making their evaluations.
The nature of the sample. Tissue samples are, by their very nature, only a small part of the whole. It is rare for a tissue to exhibit a particular characteristic uniformly throughout the entire tissue.
The subspecialty involved. Concordance for surgical tissue is so poor, for example, because the breadth of the domain is so large. There are literally millions of factors that could be involved in evaluating a general surgical slide. By comparison, evaluating breast tissue is a more focused domain, meaning that there are fewer factors to be considered.
Humans being humans. There is a natural variability in people. They have varying levels of talent and experience. They notice different things. They have different perspectives on what is important. They have bad days. They get distracted. They make mistakes. All this leads to a natural variability in the way pathologists will evaluate a particular tissue slide.
Digital pathology is often held out as a way to remove some of the subjectivity in analyzing tissue samples and providing a clinically useful diagnosis. This “computer-assisted pathology” provides pathologists with a quantifiable and consistent foundation for interpreting the findings in a tissue slide. This can be effected in several ways:
Image normalization. Digital processing of pathology images makes it possible to adjust images to account for over- or under-staining of the tissue. This provides a consistent reference for pathologists and most importantly computers examining a slide.
Algorithmic consistency. Give the same tissue sample to six pathologists and ask them to determine the percentage of cells that meet a given criteria, you are going to get six different estimates. Computers are much better than humans at precisely counting items in large sets. Ask an algorithm to determine the same percentage, and it will give you the same answer every time. This consistency ensures greater predictability.
Second opinions. Once an image is captured by a digital pathology system, it can easily be shared with other pathologists, anywhere in the world. While a physical slide is difficult to share, slide images are simple to share. This reduces subjectivity in two ways. First, it is easier to get the right pathologist to look at it. Studies have shown that there is greater concordance among specialists in a particular tissue than there is among more general pathologists. Second, it becomes easier to share with multiple pathologists. If two or more pathologists give the same opinion, it will increase confidence that a correct diagnosis has been reached. Or, conversely, if there are differing opinions, that increases confidence that more investigation is needed, avoiding unnecessary clinical intervention.
Integration of outside data. Much of pathology has depended on the facts visible in the tissue as the sole factor in determining a diagnosis. Computer-assisted pathology, or computational pathology, enables the integration of the slide images with information from a patient’s history, DNA sequencing, and other molecular tests, providing pathologists with a more complete picture of what is taking place in the tissue.
Pathology will always remain a highly subjective medical specialty, but the more subjectivity we remove from the equation, the more we will achieve greater concordance on the correct diagnosis and, ultimately, better patient outcomes. Digital pathology, with its use of image processing techniques, offers a first step in using computers to provide pathologists with a more objective assessment of a tissue. Computational pathology, by integrating information from different sources with the digital images, and by incorporating machine learning and artificial intelligence techniques, promises to remove uncertainty.