In my talk entitled “Artificial intelligence in eye health care” I give a brief history of retinal imaging and diabetic retinopathy. These two should provide the perfect use case for artificial intelligence to support the ever-growing burden of diabetes. Yet, artificial intelligence has not found its way into the mainstream screening and care of diabetic retinopathy. I highlight the main obstacles that are often misrepresented.
The 17th annual Science and the Parliament event concentrated on Science, Innovation & the Economy.
Below my talk, which was part of the panel discussion on Science & Industrial Strategy where I was given the question: Do we have the right policies in place to foster innovation that leads to economic growth?
A staggering number of reports and statistics try to answer that question. I have read quite a few and here is my two pennies’ worth.
First I will present my assumptions. Innovation is an overloaded term and here I mean technical innovation, which is by many considered an important route to grow the economy. Innovation needs investment, and this is measured in GERD and BERD, which broadly means government investment and business investment respectively. In Scotland BERD is and has been very low compared with OECD and EU28 countries and this is not improving. It is reported as a key indicator by the Scottish Government. What can we do to increase BERD in Scotland?
Let us dig deeper into statistics reported for Scotland as per 2015.
So, if BERD mostly happens in large businesses through non-government funds then a less abstract question is: how does Scotland get more large business and how do we get these businesses to invest in R&D?
My first proposal is to add larger venture capital investments for growth of innovative companies. In 2014, the Royal Society of Edinburgh reported in on the lack of growth capital for businesses in Scotland. Leading innovation countries such as Germany and the Netherlands are adding long-term investment models. The government grants capital to investment funds who in turn invest risk capital in innovative entrepreneurs. In 2005, the Netherlands started the Seed Capital programme where business growth can be supported up to twelve years. After running for ten years this policy has been so successful that in 2015, the annual input from the government was equal to its annual earnings from previous investments as a result of many businesses that had flourished. Scotland pioneered a similar model through its Scottish Investment Bank, which started in 2003. Although it invests more than the Dutch Seed Capital per annum, we have neither seen this translate into a significant increase in large companies nor in BERD. One explanation may be that larger individual investments and longer commitment are necessary to sustain the growth needed to a become a large company.
My second proposal is to invest in skills relevant to technology innovation. Students graduating from universities and colleges should be equipped with skills to grow successfully the businesses they work for or start themselves. At this point I can start talking about vague concepts such as entrepeneurship, but I will not. Instead I will focus on only one skill that is key to the successful exploitation of technology innovation: the ability to protect intellectual property. I refer to the low number of high-tech patents filed from Scotland. A number that is several times lower then high-performing areas in the EU. It becomes more than ten times lower when comparing with leading areas such as South Sweden, Helsinki and Stockholm. Universities and colleges must prepare students for the simple fact that intellectual property must be protected.
The journey to a growing and inclusive economy is more complicated than these two pennies’ worth. One fact is that Scotland has decided innovation is one of its four key drivers to a successful economy. Therefore we must ensure the people who are to undertake innovation are adequately funded and skilled.
With a small team comprising members from different companies and representatives of the American Academy of Ophthalmology, a new supplement was delivered from conception to approval in 14 months.
Supplement 173 defines Storage SOP Classes to enable anatomically correct measurements on wide field ophthalmic photography images.
Vendors have implemented new technology that enables the acquisition of OP images using wide field fundus photography. The Ophthalmic Photography IOD does not address wide fields, varied pixel spacing, and proper measurement of a stereographic projection or other methods of projection/mapping. Since the back of the eye is approximately a concave sphere, taking a very wide field image of it introduces large errors in any attempt to measure a lesion in that image (the error is very large when using a single value for the DICOM Pixel Spacing Attribute.). Therefore, DICOM WG 9 (Ophthalmology) has determined that two new Information Object Definitions (IODs) are necessary to adequately represent wide field fundus photography.
Manufacturers of ophthalmic photographic imaging devices have been developing OP images (using a narrow field) for many years in DICOM (i.e., these SOP Classes are widely supported by the DICOM ophthalmic community). Therefore, the wide field OP image storage SOP Classes are an extension to already existing narrow field DICOM SOP Classes.
Medical imaging acquired for clinical purposes can have several legitimate secondary uses in research projects and teaching libraries. For these uses, and when imaging is acquired directly for research, most personal data can be safely removed, but in many cases it is necessary to keep some personal data or a link with the personal data. At this moment, no commonly accepted solution exists because the amount of personal data that is required by researchers and teaching libraries varies case by case.
DICOM Confidential is an open source DICOM (Digital Imaging and Communications in Medicine) de-identification toolkit that provides the necessary flexibility to account for different de-identification requirements and does not impose a given anonymisation model. It also provides a mechanism for forwarding the anonymous output to a remote site using either SFTP (SSH File Transfer Protocol) or DICOM communications protocol. The core functionality is contained in a Java library, which we have used to develop two DICOM anonymisation applications; these are included in the toolkit: one for DICOM files contained in a folder and one for objects received via the DICOM protocol (a receiver). A separate graphical application is provided to help users in policy writing and configuration.
In the case of multicentre clinical research projects a uniform, accountable de-identification process at all the participant centres is desirable. This can be done by prior distribution of the Privacy Policy to be used or by deploying it in a web server from which DICOM Confidential can read it directly. The toolkit was deployed in a multicentre setup and the experience gained in this deployment is presented.
Get it from SourceForge
Rapid is a cost-effective and efficient way of designing and delivering portal interfaces to tasks that require remote compute resources. The aim of Rapid is to make completing these tasks as simple as purchasing a book or booking a flight on the web.
The philosophy of Rapid is to deliver customised graphical user interfaces that enable domain specialists to achieve their tasks. These tasks make use of domain-specific applications that run on remote compute resources; a requirement which is satisfied by translating the task into one or several computational jobs to be performed on Grid and Cloud Computing infrastructures, and High-Performance Computing facilities.
Customised interfaces allow tasks to be performed without referring to terminology about the underlying computational infrastructure. Moreover, the system allows to expose particular features of applications as not to overwhelm the user.
Contains a collection of source files authored by different people that put together allows the extraction of features from travelling salesman problem instances.
See these papers for a detailed description: