Digital pathology and Artificial Intelligence in the precision medicine era
Posted 15th September 2017 by Jane Williams
In 2017, technological advances in cloud computing and artificial intelligence has pathology positioned to become one of the most talked about medical fields in healthcare.
Pathology, a field with little direct patient interaction but instead obtains information based on the pathologist’s unique interpretation of tissue as seen under the microscope, has long been overshadowed by other more front-facing medical specialities like surgery, oncology, or radiology. While human genome analysis has generated a great deal of well-deserved buzz in moving us towards precision medicine, a great deal of disease remains defined by what can be seen under the microscope. Pathologist involvement in precision medicine and digital technology will be a key driver for improving patient care.
The adage that an image is worth a thousand words is apt in describing the utility of digital pathology for precision medicine. Gigabyte sized tissue images provide a wealth of information encompassing the morphology of disease. Not only can one visually identify the presence of disease, but also the spatial interactions between different types of cells, blood vessels, tissue compartments. This has never been more important than now with the development of immunotherapies where understanding the interplay of factors within the context of the tissue microenvironment will determine whether treatment is effective. Maximising the potential of such a rich data source has huge implications for healthcare and digitalization of pathology data will be the first step towards improving patient outcomes.
The benefits of digital pathology over traditional pathology are manifold, including non-destructive analysis, backups and instantaneous sharing of information across sites. Mass digitalisation of slides means at least 3 things.
Firstly, other pathologists can review and communicate knowledge by simultaneous viewing of the tissue. This has already been put into action as pathologists have created platforms to upload and share annotated pathology data for the purpose of educating pathologists at smaller institutions, particularly on rare case studies. In addition to rare case reviews, diagnosis of various immunohistochemical stains only continue to get more complex as more assays are developed and connecting appropriate experts to consult on difficult to interpret stains will ensure better outcomes.
Secondly, algorithms can be used to analyse the image and provide computational decision support systems for pathologists and provide a scalable method to connect them to various diagnoses and treatment options. Modern cloud computing infrastructure has been absolutely vital to the success of digital pathology as a platform. Once the slides are in the digital domain, additional workers can be summoned on demand to provide relevant information using image analysis algorithms or querying a database of similar cases to help the pathologist make a more accurate and confident diagnosis.
Lastly, patients will be able to see and understand their disease in the same way that we see a fracture on an x-ray or a lesion in a mammogram. The ability to visually convey the diagnosis increases trust between the doctor and patient and can provide for meaningful conversations that would not be possible otherwise. Pathologists have a great deal of knowledge about disease to share and we will see the visibility of pathologists increase as digital pathology provides a greater platform for showcasing their work.
The marriage of artificial intelligence with digital pathology will be the key to realising the potential of personalised healthcare. Major advances in image recognition and deep learning have allowed pathology images, a highly descriptive source of dark data, to be mined automatically and at times more accurately than humans. Data reproducibility is a major concern and digital analysis has been shown to cut down on inter and intra reviewer variability. Automated algorithms will allow for more complex analysis of cell counts and cell-cell interactions for developing very specific image-based biomarkers. Bringing AI into pathology can provide almost limitless amounts of data by scouring information simultaneously at the tissue, cellular, and subcellular levels.
While there are concerns about artificial intelligence replacing the role of pathologists, the reality is there is too much at stake to simply leave diagnosis to computers. Given that many studies have shown better outcomes when decisions are made by combining the knowledge of pathologists with computational support, artificial intelligence will be an invaluable tool for reduce the variance between different pathologists, free up time spent on more routine tasks, and allow them maximise their expertise to be more accurate and confident in their diagnosis.
Connecting unprecedented amount of data to an unprecedented number of treatment options will necessitate computational power to analyse the data. The CEO of Google has recognized pathology has a core focus for AI research and digital pathology has been a focus of several other exciting ventures. It is clear that we will see massive gains in computational decision support systems in the years to come.
While there remain many barriers in setting up the environment for routine digital pathology integration, the need to improve healthcare outcomes at scale should ensure that pathology images are included in the digital workflow at more institutions over the coming decades. With pathologists leveraging slide sharing technology and artificial intelligence to better educate the public and each other, a digitised pathology world promises to make the pathologist more prepared, more recognizable, and more valuable than ever before.
George Lee is Digital Pathology Informatics Lead and Data Scientist at Bristol-Myers Squibb.
Interested in learning more about the role of digital pathology? Download the agenda for the upcoming 4th Digital Pathology Congress.
Leave a Reply