Very interesting post related to thoughts over many years and image analysis courtesy of Jim Diamond over at PathXL’s blog in their piece entitled “Has Image Analysis made it into Modern Pathology?” or what should have been called “John, you’re a Timex watch in a digital age”.
Has image analysis made it into modern pathology? Read below for one approach and answer to this question.
Many years ago I had experiences with two senior UK based pathologists, both of which took place at major national conferences.
At a conference many years ago I was trying to convince a senior pathologist that a new kid on the block – Digital Pathology – had something to offer within his discipline. To this end I endeavoured to show him the PathXL content management system. When I logged in and was prompted with the list of virtual slides on screen, he immediately responded in a critically defensive manner by explaining to me that there “is just not enough information on the thumbnail for me to make a diagnosis”. He suggested innovatively that if I could make the image bigger this may produce an acceptable result.
My second experience was with another pathologist at a difference conference. I was extolling the virtues and merits of using virtual microscopy in a remote way to deliver QA programmes across the UK. He was routinely sending slides in boxes to participants via snail mail. His unforgetful response to a computer science researcher was not that original but typical of older pathologists – “I’ve always done it this way and you should be aware that computers are just going too far these days”.
You’re a Timex watch in a digital age
Thinking of these pathologists I couldn’t help but think of the protagonist in the film “Die Hard 4” who when referring to John McClane said, “John, you’re a Timex watch in a digital age“.
Thankfully these days are gone and pathologists are now looking forward to a future that sees computational techniques merge as an adjunct to their work and a technology that empowers rather than threatens their role in medicine.
“A large-scale switch to digital pathology systems was imminent and that Evolving technologies and market forces reveal that digital pathology is poised to radically affect the daily workflow and activities of pathologists and diagnostic laboratories”.
(Frost & Sullivan Report – US & European Digital Pathology Market – 13 June 2013)
These comments are now backed up in the concept of “The birth of the Computational Pathologist” a system that can operate as a pathologist. I prefer a literal interpretation of this phrase as more-and-more often we see the pathologist being drawn into computational processes, imaging modalities and data mining techniques.
It is generally believed that the significant innovations in pathology diagnostic systems are those that automate/reduce manual intervention previously performed by Biomedical Scientists and MLAs. These are categorised under many modalities including, sorting tubes, de-capping, centrifuging, sectioning, tissue mounting and staining. These are all seen to:
- optimize workflow
- improve productivity
- increase preparation consistency for enhanced diagnostics.
However, these are processes that address the current state of laboratory operations and consequently most of the automated solutions surrounding digital pathology involve implementations in one or more of the following areas:
- Integration: Integration with current operational LMS/LIS.
- Organization: Management of pathology workflow within a laboratory.
- Observation: View your cases from an intuitive process management system.
- Consultation: consultation with local or remote pathologists/scientists.
However there is a further category that requires implementation if Digital Pathology is to claim the “Breakthrough Technology” crown and not be a digital implementation of the current technologies. This category is that of diagnostic automation whereby a computer can intervene in the diagnostic process.
Digital Pathology is often described even today as the scanning of a glass slide into a whole slide image; yet, it is so much more. It will become a revolutionary tool when it addresses the issues of “Computer-Aided Diagnosis (CAD)” as routine. This modality is rapidly entering the radiology mainstream. It has already become a part of the routine clinical work for the detection of breast cancer with mammograms. It is however not a well recognised modality in pathology but it must be the goal of digital pathology solution providers to ensure that this is implemented.
When PathXL was founded through spin-out of scientific research in pathology from Queen’s University Belfast (QUB), the prime mover here was to introduce novel, innovative and scientifically relevant advances into pathology. The founding product of PathXL was InView (now known as PathXL Simulator) which introduced pathology to some of the research going on in decision support systems and how pathologists made decisions. These tools were some of the first applications to make use of the early stage Virtual Slides. In fact these digital slides were the enabling technology that allowed InView to be effective.
It was clear even as far back as 1998/99 that the concept of the virtual slide was going to revolutionise image analysis and its role in automated diagnosis. Until that date image analysis had always been a “proof-of-concept” notion. It had been shown that image analysis applied to small images could highlight morphological characteristics of both tissue and cellular components and relate these to disease progression. This missing link was to be able to take image analysis to the next stage and produce an image analysis tool that could be diagnostic in nature.
All nice thoughts back 14-15 years ago but where are we now?
Has image analysis made it into modern pathology?
I would argue not to any major extent and certainly not to the extent that radiology has embraced automated intervention. Image analysis has not made it into routine diagnostic histopathology. All image processing is being directed at the research arena. Elements of image analysis have entered the cytopathology arena where systems such as AutoPap are employed for the detection of cellular abnormalities in cervical screening. This system has been shown to speed the examination of Pap smears without lowering the detection rate of clinically important lesions, thus helping alleviate the cytotechnologist shortage. Studies have shown that it can double through-put.
Cytology based systems are essentially object detection (rare event) systems, where there is no real consideration given to context. The inherent architecture in histopathology and the variants of it are a significant problem. Accordingly, diagnostic histopathology does not currently use any automated diagnostic systems. This is the probably the Holy Grail for computational scientists in pathology as it is certainly the most complex.
In the research arena many vendors are now applying image analysis to histological imagery; this is usually in the quantification of Immunohistochemistry (IHC). But, the required need now is to provide tools that can quantify virtual slides that are routinely stained with Haematoxylin & Eosin (H&E) for tumour. A qualifying factor is that these tools need to be integrated into the pathology workflow so that analysis can be achieved in an efficient way where 1000s of slides can be analysed.
Pathology solution providers today now need to make sure they do not get into the role of developing “Timex Watches” or solutions that don’t really deal with the forward thinking demands of pathologists. PathXL currently has a strategy to move firmly into this domain of tissue analytics bringing years of academic development to the commercial world with the desire to achieve “Excellence in Tumor Analysis” through the provision of effective imaging solutions.
To find out more about the developments of TissueMark and for a free demo of our digital pathology solutions, get in touch with our expert team today. You can keep up to date with all the latest news and developments at PathXL by signing up to our newsletter or connecting with us on Twitter, LinkedIn or Google+.