Transforming medical image analysis with deep learning AI
Posted 5th April 2019 by Joshua Broomfield
Aiforia Technologies is a medical AI software company seeking to transform clinical pathology and medical research by bringing deep learning AI to assist and augment human experts in medical image analysis.
We had the chance to ask CEO Kaisa Helminen about AI in healthcare and Aiforia’s newest platform.
How is AI going to revolutionise healthcare?
Medical image analysis is one of the first areas in healthcare where AI holds the most promise. With the introduction of neural network based image analysis, it is possible to automate detection and quantification tasks that have been nearly impossible for the human eye. With AI-assisted analysis, one can reach new levels of information and data that has been “hidden” in the sample before.
Revealing all this information will allow researchers and medical professionals to start asking new questions, to make new discoveries from biological samples, and to move towards more quantitative science. It will also enable pathologists to significantly improve the efficiency and consistency of their work, while computers can be used to do part of the work for them.
I believe that AI, and convolutional neural networks, in particular, holds immense potential to make the work of healthcare providers more efficient and streamlined than repetitive. Being able to bring technological advances from other fields to the medical sector to benefit patient care is what drives me the most. We strive to bring AI to the fingertips of clinical pathologists and other medical professionals to enable better, faster, more personalised care for the patients globally. AI also has the potential to democratise healthcare while enabling rapid and accurate diagnostics support also to rural areas and developing countries.
AI enables the development of personalised healthcare services as well as significantly reduces the time needed to gather information from various sources for decision-making. It brings new possibilities for home care and patient monitoring, as well as robot-assisted surgeries. The possibilities are endless.
Besides augmenting medical experts in diagnosis, deep learning AI and the software tools what we have developed, also facilitate new scientific discoveries and development of new drugs to diseases, which affect people around the world. One of the most interesting scenes in the future is to combine genomic information to tissue morphology data, to really ask new questions from all this big information, to understand biology better than before.
I find this truly fascinating and can’t wait to start seeing new discoveries being found.
How can the Aiforia® Cloud assist pathologists in their daily work?
The name “Aiforia” derives from the words “AI for Image Analysis”, a perfect description of what we do. We launched our novel state-of-the-art deep learning image analysis platform Aiforia®Cloud last year. It is the first-of-its-kind solution in the market that enables rapid development of deep learning based image analysis algorithms, specially optimised for digital pathology applications, but suitable for any 2D images.
The Aiforia®Create module of our software provides unique self-service AI development tools for end-user domain experts. With this platform, these domain experts (e.g. pathologists) can create their own image recognition algorithms to automate their quantitative pathology or general image analysis tasks. We naturally also offer professional services for purpose-built algorithm development based on customer needs.
In Aiforia®, the image analysis algorithms are created by training convolutional neural networks (CNN) to learn, detect and quantify specific features of interest. Our ethos is that if you can visually identify your feature of interest in images, it is possible to automate the discovery with Aiforia® and quantify the features – accurately and reproducibly.
The often overlooked benefit or our Aiforia®Cloud solutions is that there is no need for users to invest in local hardware or software. Our platform can be instantly deployed and it scales easily based on demand. An additional benefit of the cloud setup is the facilitation of knowledge and discovery sharing, as well as enabling remote consultation for areas suffering from a shortage of local experts.
Aiforia®Cloud, and especially the Create module, is developed to augment the work of pathologists, and to place the pathologist in control of the newest Image Analysis tools. The well-worn slogan, ‘the aiReady professionals will replace the unaugmented professionals’, is probably true and we want to make the adoption of AI as seamless and enjoyable as we can.
With the help of Aiforia®, pathologists can improve the efficiency of their work, e.g. delegate tedious counting of objects, measuring tumor burden, or type, or identifying rare targets in samples. Pathologists will still review the results and make the final decision. With the help of AI, they can spend their time more effectively and concentrate on the more difficult cases.
A different characteristic of drug development is the need for higher dimensional data than just primary diagnosis according to classic standards. Especially apparent in the Immuno-Oncology (IO) field, there is an increasing need to quantify, at large scale, subtle spatial relationships between different classes of immune cells.
Thus, pathology in clinical research and drug development relies on very different metrics and quantitation tasks that can be quite beyond the scope of any human to perform. e.g. multispectral images with 10 or more channels combined with potentially millions of objects. All this ‘data’ is ‘information’, but this is the beginning of a new ‘Big Data in pathology’ era that gives us the possibility to think about completely new ways of experimental design and analysis.
Are we seeing the level of digital pathology adoption we would expect? What do you think are the key blockers to adoption?
The field of digital pathology has taken quite some time to mature, but I think we are finally witnessing its wider adoption. Digital pathology has been well adopted in medical research and drug development processes already, and we are finally seeing quite a lot of progress in the clinical diagnostics workflows as well towards digitalisation.
Of course, converting these complex workflows takes time and capital investment. I think that the accelerating uptake of digital pathology indicates that there is both an increasing economics argument in favour, as well as workflow and operations control argument.
I also think there is increasing recognition that we are simply putting too high a burden on our pathologist. They are a critical and limited resource of immensely skilled doctors and diagnosticians. We should really make every effort that they are able, with AI-driven augmentation, to focus their energy on the most important and interesting tasks.
Of course one of the fundamental requirements before AI can assist pathologists is to digitalise the histological samples and thus ensuring microscope scanner resources to anatomical pathology departments is a must. I also think that developing microscope scanners is crucially important to increase the speed of adoption. Scanners which will be compatible with a wider variety of different sample types, including varying resolution requirements for example, and serve all tiers of the market, not just the highest throughput laboratories.
These hurdles that still remain are ‘only’ related to economics. We have passed many of the scientific and technological hurdles already, and regulatory hurdles are being addressed on multiple fronts as we speak. We are working with the various stakeholders to design our technology so that all of these requirements are filled and surpassed.
The bright future will bring care and diagnostic help to underserved areas, higher throughput in congested areas, more personalised care for patients. It will ultimately bring down the cost of care of this particular process, which we as a society should also be mindful of. If we keep an eye on the future and remain committed to solving the technical issues that any workflow disruption will have there is no doubt in my mind that we can enable better patient care and quality of life.
That’s what it takes and early adopters are going to be thankful they got ready when they did, as will the patients they care for.
Kaisa Helminen is CEO at Aiforia Technologies Oy. Aiforia® are Diamond Sponsors at the upcoming5th Digital Pathology & AI Congress: USA and will be delivering a pre-conferencing seminar on their deep learning image analysis platform.
The 5th Digital Pathology & AI Congress: USA is on June 13th-14th. Company showcases and interactive demonstrations of AI tools are a feature of a packed agenda. This meeting has a reputation for providing an outstanding networking experience, so register now to avoid disappointment!
Leave a Reply