ECR_2016

EUROPEAN HOSPITAL  Vol 25 Issue 1/16 EH @ ECR Robust statistical evaluation depends on the source image Give a computer enough image data and the algorithm assesses details Big data and computer- assisted diagnoses Deep learning software One thing is certain in big data discussions: Intelligent machines will change our world considerably. What is less certain is exactly how these changes will look. Although networked data processing offers many opportunities, its development is still in the early stages. In medicine, there is great hope that it will be possible to extract and use valuable information hidden in the masses of digital data in a meaningful way. Those currently involved in the research and development of innovative technologies are likely to be among the big winners resulting from this boom. Here, Peter Aulbach, Marketing Manager at Siemens Healthcare, gives us some insights into his company’s present and future strategies around big data In a recent skin cancer study, computer software out-performed experienced doctors by such a wide margin that Harz says: ‘I wouldn’t have to think twice about which diagnosis to trust more’ There has been some significant suc- cess in computer-assisted detection. CAD systems, such as those used for mammography and lung screening, look for typical patterns and recog- nise irregularities that might indi- cate pathological changes. ‘The soft- ware marks the conspicuous parts of the data sets at which the user should take a closer look. The objec- tive is not to replace radiologists but to support them in their work, so that they do not overlook anything. Doctors will continue to make the diagnoses,’ Aulbach points out. Things should become even more exciting once these artifi- cial intelligence learning software programmes can make prognos- ‘In the Automation in Medical Imaging (AMI) project, we will build the necessary tools and infrastruc- ture to ease the development of a special kind of computer software,’ computer scientist Markus Harz explains. ‘This software is capable of learning, so that it can at some point understand images as well as a human. As the project name implies, we are not trying to teach the computer to understand any image, but medical images in par- ticular. This requires infrastructure: loading clinical image data into the computer software development process is not straightforward, and making standard desktop comput- ers learn is cumbersome. This infra- structure helps to develop the soft- ware efficiently. ‘AMI is an international research project jointly headed by Horst K Hahn and Markus Harz between Fraunhofer MEVIS in Bremen and the Diagnostic Image Analysis Group (DIAG) in Nijmegen, The Netherlands.  The DIAG group has renowned expertise in state-of-the- art self-learning software for med- ical image analysis. Fraunhofer MEVIS has years of experi- ence in industry-grade soft- ware development.’ The com- bination has already proved complementary, he points out: through shared learn- ing and fusing development systems. Scientific, commercial, technical objectives ‘The greatest goal is to create software that solves real clini- cal problems. We are convinced that automation helps. Automate the processes clinicians hate most, like searching for the right images, or comparing details from two exami- nations. Clinicians will want the software, thereby increasing mar- ketability, so we’ll have reached our clinical and commercial goals. Self-learning software ‘This software is similar to a per- son who learns. When a radiologist decides whether a suspicious area in a medical image is harmless, or reason for concern, she will use all her experience and knowledge, to compare form and structure, perhaps making judgments based on location and assessing other fea- tures. This is very similar to various approaches taken by machine learn- ing. More traditional methods use criteria such those used by radiolo- gists. The magic lies in teaching the software to “see” with a radiologist’s eyes and judge by her criteria.’ Deep learning algorithms ‘Deep learning comprises a vari- ety of neural network architectures. Neural networks emulate the brain’s neural connectivity with neurons and synapses. Deep-learning neural networks are special among these approaches. They contain many more neurons and synapses than previous neural networks. Perhaps most significantly for machine learning, deep neural networks can learn directly from data. Previously, experts had to translate between image and learning algorithms: They designed the image features for the computer to use. In a way, an expert taught the computer to see. Deep learning is different, giving the computer huge amounts of data and hints about the meaning, so that the deep learning algorithm can discover the most relevant image features  – often much more useful than those constructed by a human.’ Areas of use ‘We have three clinical applications in mind, but the deep learning approach imposes a clear restriction: it requires large amounts of data, and this data has to be prepared. The project tackles this problem: We want to ease the development of deep-learning computer software by simplifying data preparation by doctors, to collect large amounts of data quickly. Thereafter, trained computer software helps to collect more data. A virtual circle! We begin by focusing on three applications for which substantial data already exists –  digital pathology, ophthal- mology and oncology.’ ‘The objectives vary for each of these fields. Oncologists moni- tor the health of cancer patients in screening or after treatment. They search for minute, s u s p i c i o u s changes, but the images are most often completely unsuspicious. We want to present these tiny chang- es to the oncolo- gist even before he’s viewed a single image. The software must learn about the body and organs, how they normal- ly appear, where they are. Finding and capturing them automatically greatly helps later processing. ‘For digital pathology, the chal- lenge is to manage extraordinarily large images. The search for tiny cancer cell clusters resembles seek- ing a needle in a haystack. This is exactly what a computer can do very well and very accurately. ‘For ophthalmology, we want to improve treatment decision-making, for example, by teaching software to detect and measure fluid compart- ments behind the retina. The change of this exact value from exam to exam determines the treatment.’ Software diagnostic advan- tages/disadvantages ‘Computers can now match clini- cians’ performances, at least for isolated tasks. The most prominent examples include tumour detection in breast and lung cancer screening and diagnosis of skin cancer imag- es. However, the algorithms can only perform well when provided with sufficient data and informa- tion. Doctors often have an informa- tion advantage: they may know the patient from a prior visit, or read a relevant journal article, or dis- cussed the topic with a colleague. Computers cannot easily access such implicit knowledge. Therefore, I think humans will certainly be part of the picture for quite some time.’ Endangering medical jobs ‘The help of ‘intelligent’ computer software in medicine is a means to extend and improve healthcare. I see radiologists struggle with the sheer amount of medical images they must review. How long can this continue? Employing more radiologists seems unfeasible, given already explod- ing costs in high-tech medicine. Simultaneously, public awareness of the benefit of modern, image-based diagnostics increases the demand to offer this to a broader public. This is possible, but hardly imaginable tic statements. Siemens is currently testing software that can differenti- ate between threatening and non- threatening coronary stenosis. The analysis programme deter- mines the virtual blood flow reserve (CT-FFR) in the coronary ves- sels during a cardiac CT scan and decides whether a relevant stenosis is present, or not. In the future this could avoid unnecessary cardiac catheterisations. What is hardly ever mentioned in the context of the algorithms on which these learning programmes are based, says Aulbach, is image quality: ‘Achieving significant results for the analysis and processing of data requires perfect raw data. Image acquisition therefore calls for the utmost care and precision, as there may be artefacts that later can- not be eliminated. The robustness of statistical evaluation methods there- fore depends on the source image.’ The marketing manager does not see a problem in contrast media administration, which concentrates in a different way in each individual patient and consequently produces different image information: ‘Dual- energy-CT facilitates mono-energetic CT imaging, which automatically balances the different intensities of contrast media concentration in different image data sets. For the purposes of comparative analysis, it’s therefore not relevant whether some data sets have higher contrast media concentrations than others.’ The big data programmes cur- rently in use are only aimed at assisting doctors. However, at some point, and with the help of learn- ing computer systems, there is a chance that diagnoses will become safer and faster – and all without the human factor. This is a topic in which Siemens is also interested. Currently, one of the main objec- tives is the development of a fun- damental telecommunication base because comprehensive data pro- cessing necessitates feeding the machines with information. As a global manufacturer of medical technology, Siemens has access to masses of usable data. However, utilising this data requires the cus- tomers’ consent. Currently, the com- pany is developing a cloud-based network entitled ‘Teamplay’, which not only is used but also ‘fed’ by doctors, clinicians and other health- care providers. Customer data is anonymised, collated and processed, based on certain patterns. ‘One of the first applications we will offer with this IT platform is the optimisation of dosing protocols,’ Peter Aulbach reports. ‘We derive and evaluate pat- terns from the dose values supplied by the users. ‘We then replay the results to the customers. This enables larger hospitals, for instance, to check whether their CT scanners adhere to the threshold values in all locations. Even more interestingly, it also allows different centres to measure themselves against one another. If a hospital is hoping to be among the world leaders when it comes to curbing radiation doses it can use our system to check what it should do to keep up with the ten best cen- tres in this field. ‘Quality assurance, such as found in the manufacturing industry for years, could then finally become reality in the healthcare sector as well,’ Aulback concluded Peter Aulbach is a marketing manager at Siemens Healthcare Deep learning algorithms autonomously find interesting spots in new digital images of tissue samples based on an automated analysis. Starting with the highest resolution, these neuronal networks compress the data until information and image interpretations emerge

Pages Overview