This week saw Frost & Sullivan award its 2013 Company of the Year to Definiens, a provider of image analysis solutions and data mining solutions for life science, tissue diagnostics, and clinical digital pathology. Definiens’ gaining of the title owes much to its work around tissue datafication that’s leveraging its Definiens Cognition Network Technology, which the company says mimics the human mind's cognitive powers to reposition knowledge within a semantic network.
“What we do essentially is look at ways to be able to better diagnose cancer and develop therapies,” says Merrilyn Datta, CMO at Definiens. The company looks to extract data from tumor images, historically available as slides from biopsies, datafying the tissues involved to create digital images and then using its Cognition Network Technology to extract all the different relevant objects in that image and correlate them to patient outcomes. “That can be extremely, extremely powerful,” says Datta.
The image analysis technology was developed by Physics Nobel Laureate Gerd Binning, and includes a set of principles aimed at emulating the human mind’s cognitive powers, which are defined by the ability to intuitively aggregate pixels into ‘objects’ and understand the context and relationships between those objects rather than the computer’s normal way of just examining images pixel by pixel. These principles include: context, which is established and utilized through the technology’s creation of a hierarchical network of pixel clusters representing nested structures within the image; navigation, for supporting efficient navigation inside the network in order to enable specific local processing and addressing of specific contexts; and evolution of the network, in which the individual stages of segmentation and classification are alternated and the structures represented within the network are created and constantly improved in a series of loops, whereby each classification can be enhanced with local context and specific expert knowledge.
As Datta explains it, in traditional pixel by pixel image analysis you might see a pixel lit up to a certain intensity and conclude something is there. “But cognitive network technology essentially lets you create an object and a context, so that pixels are actually distributed to an object that you define,” she says. “Say that I want to look at these objects that are circular and this color next to these other objects that are blue and finer-shaped. The software can look across the slide set, for how many times two objects are seen in that context and then [data-]mine to say that where everything is seen in this context, the outcome for that patient was good when we applied this medication, or the outcome was such-and-such in women of this age. So it’s about making correlations when objects are found in a certain context.”
The dataification process involves leveraging machine learning to help train the system to quickly configure the detection of regions of interest, such as what circular objects look like. The software also supports metadata handling around things like patient information, and users can make correlations between the datafied image (which is both visually and numerically-defined) and the metadata to draw conclusions about populations, for example. Today, its digitization and datafication can help in determining how a woman with breast cancer will respond to certain treatments if and to what degree a protein called human epidermal growth factor receptor 2 (HER2) –positive is present in cells in tissue images. “But the future is going to be about using full dataification of all the objects in an image to understand all possible features that might be important in cancer and discover new biomarkers and make new cancer tests,” says Datta. “That’s on our roadmap for where we believe the power of tissue analysis will go.”
Definiens, Datta says, is a market leader in image analysis in drug discovery in the pharma sector, but in recent years it’s moved to the clinical arena, “which is looking at actually how to use this to stratify patient populations,” she says. “Where we see the final ultimate utility is for personalized medicines and stratifying patients’ therapeutic decisions. We are at the very beginning of that journey.” The next steps for Definiens, Datta says, is to grow Cognition Network Technology’s robustness in terms of tissue phenomics to move to the next level involving custom-crafting solutions and doing that in a more scalable fashion.
The longer-hike path going forward will involve taking on Big Data and bringing together disparate information formats, she notes, so that doctors will be able to serve up a bunch of data all at once with correlations outlined rather than the current serial diagnosis process.